# RiskRouter: every page, as text > Built from the site itself on 2026-11-17. The short map is https://riskrouter.eu/llms.txt. ## RiskRouter: prove what you decided, and that nobody changed it URL: https://riskrouter.eu/ Tamper-evident records for finance, AI, security and any decision you may one day have to defend When a decision is challenged, show the record. Then prove nobody changed it. A complaint, an inspection, an audit, a claim, a lawsuit: each ends with the same three questions. What did you decide or advise? When? And has the record been touched since? A record kept in a system you control answers the first two on your word alone. RiskRouter answers all three with evidence anyone can check themselves, without asking us, and without your client’s data ever leaving your system. Seal a file now, in two minutes Get a free sandbox key for your software No form, no contract, no card. Seal a file from this site in two minutes, or your first record from your own software in one request. The ledger right now: entries chained, head , . Check it yourself The last three entries of the live ledger as signed on 13 September 2026 and timestamped in Bitcoin block 967830. Each entry carries the hash of the one before it. #44 sealed record prev 8205f994 965a4c hash 8688a6a0 b50303 #45 sealed record prev 8688a6a0 b50303 hash c06ca6e2 4ff694 #46 sealed record prev c06ca6e2 4ff694 hash f95974f5 d8c225 Change one character of #44 and its hash changes, so #45 no longer points at it, and so on to the end. Try it . 13 September 2026 first head anchored 967 830 earliest Bitcoin block none yet independent witnesses Computed from the anchors and witness keys in the repository when this page was built. Why these cannot be faked 1 request To seal a record, and one more to fetch its proof 0 personal data Leaves your system: we receive a salted fingerprint Offline Verification needs no account and no call to us 6 laws Record templates mapped article by article Built first for the software two kinds of firm already use One integration in the software a firm already runs reaches every firm on it. That is where we start: the evidence stays in the product they know, and the record never reaches us. Software for regulated finance Broker, insurer, investment-advice and lending platforms. Their customers must keep the advice or decision record (IDD, Article 20; MiFID II, Article 25; Consumer Credit Directive, Article 18) and may one day have to show it was never changed. AI-governance and deployment platforms For high-risk AI systems, the logs a system generates must be kept by the provider and by the deployer for at least six months (AI Act, Articles 12, 19 and 26(6)). Sealed logs show they were not rewritten once a decision was contested. Not software? Seal a file by hand. Another field? Every industry . Who it is for: anyone who may one day have to show a record was never changed Most of these people are required by law to keep a record. None of them is required to prove it was never edited, until the day someone contests it. The log takes the fingerprint of any record, so the list below is where we have looked, not where it stops. You are What you may one day have to show What changes with RiskRouter Financial services An insurance broker or agent Why you recommended this policy to this client, years later (IDD, Article 20) Each advice note is sealed the moment it is written. In a dispute you hand over the note and its proof, and the other side checks it without trusting you or us. For brokers An insurer, MGA or compliance lead Advice, incidents and reports under IDD and DORA, as they were on the day Records sealed as they happen, with signed heads, two independent time-stamps and a Bitcoin anchor: a date nobody can move, including you. For insurers An investment firm or financial adviser Why a product was suitable for this client, and what the suitability report said (MiFID II, Article 25) Each suitability assessment sealed as it is made. The MiFID II template maps each field to its article. For investment advice A lender or credit platform How a creditworthiness assessment reached its result (Consumer Credit Directive (EU) 2023/2225, Article 18), and the logic of an automated decision (GDPR, Article 22) Each assessment sealed with the inputs and the rule version it used, so the decision can be shown exactly as it was taken. For lenders A bank, payment or crypto firm That customer due diligence was done, and its records kept for five years (Directive (EU) 2015/849, Article 40); what happened in an ICT incident and when it was reported (DORA, Articles 17 and 19) Each completed check and each incident report sealed when it happens. The customer’s file stays in your system. For AML and KYC A software vendor serving any of the above That the records your product keeps for hundreds of customers can be relied on One integration gives every customer a record they can show a supervisor, an auditor or a court. You keep the customer, the data and the interface. The vendor case Any organisation A team running AI or automated decisions What the system decided, and that its log was not rewritten afterwards (AI Act, Articles 12, 19 and 26; GDPR, Article 22) Drop-in recorders for your existing logs, LangChain and OpenAI Agents, and an MCP server. One sealed record per decision or agent run. For AI decisions An HR team using software to screen or rank candidates How each candidate was assessed; AI used in hiring is high-risk under the AI Act (Annex III, point 4) Each screening decision and each human review sealed as it is made, with the version of the tool that made it. For hiring A security or IT lead What your logs said before an incident, and when each notification went out (NIS2, Article 23; GDPR, Article 33) Logs sealed segment by segment as they are written, so an edit made afterwards is caught. Sealed security logs A data protection officer That every personal-data breach was documented, and what was known when (GDPR, Article 33(5)); that you can demonstrate compliance at all (Article 5(2)) Breach records, impact assessments and changes of consent sealed when written, with no personal data sent to us. For data protection Beyond financial services A software maker What you shipped, its bill of materials, and when you learned of a vulnerability (Cyber Resilience Act, Regulation (EU) 2024/2847) Register a signed statement for each release through the IETF SCITT interface; the artifact never reaches us. For software makers A pharmaceutical, biotech or testing laboratory That results and batch records were not altered after the fact (EU GMP, Annex 11 on computerised systems) Seal each result’s fingerprint as it is created; the data stays in your systems. For laboratories A company publishing a sustainability report The figures behind it as they stood when they were reported and assured (CSRD, Directive (EU) 2022/2464) Seal each dataset and each version of the report, so a later correction shows as a correction. For sustainability reporting A researcher, designer or inventor That a document, design or dataset existed by a certain date, without publishing it Seal its fingerprint. The content stays private, and the date rests on public time-stamps and Bitcoin, not on your word. For research and inventions A legal, company secretarial or finance team Which version of a contract was signed, and what the board resolved and when Seal each version when it is agreed or signed; anyone can later check it is the sealed one. For contracts and boards A contracting authority Each step of a procurement procedure, enough to justify every decision (Directive 2014/24/EU, Article 84) Seal the scores and the evaluation report as each step closes, so they cannot be quietly adjusted after a challenge. For public procurement Anyone else A quality inspection, a policy in force, a record nobody has thought of yet The same call, a different kind , or a file sealed by hand. Every industry The quickstart Record templates exist today for IDD, MiFID II, the AI Act, the GDPR, consumer credit, anti-money laundering, DORA, NIS2, the Cyber Resilience Act, VAT invoices and public procurement. For everything else the log already takes any record you send; there is simply no field template for it yet. Sealing a record makes it provable, not compliant: what a law requires you to keep is still yours to judge. Why a record in your own system is not enough Anything that can store a record can also correct it. That is not an accusation, it is how databases work, and it is why a supervisor, a mediator or an insurer cannot tell an accurate history from a tidied one. The question is not whether you keep records. It is whose word they rest on. Your own database or archive A supplier’s audit report RiskRouter Who can change a record later Anyone with write access, often without a trace an outsider can check The supplier who runs the system Nobody, including us: one changed entry breaks every entry after it, against heads already published Whose word the proof rests on Yours The supplier’s Nobody’s: open-source verifiers, public keys, Bitcoin Where your client’s data goes Stays with you To the supplier Stays with you; only a salted fingerprint is sent If the supplier disappears Not applicable The report goes with it An export you hold still verifies, offline, without us Compared with the other ways to prove a record’s date Time-stamping services and ledger databases are good tools, and a buyer should weigh them. This is where each one stands, including the row where they are ahead of us today. A time-stamping or notarisation service A ledger database RiskRouter What leaves your system A hash of each document Usually the record itself, in their database or one you run A salted fingerprint, never the record What it proves That one document existed at a time That the history inside that database was not edited That each record existed unchanged from a moment, in a public log nobody can edit Who can check it Anyone with the authority’s certificate Usually the vendor’s own client or service Anyone, offline, with open-source verifiers in JavaScript and Python and a published specification Legal weight of the time Can be qualified under eIDAS, with a legal presumption of accuracy None of its own Not qualified today; the time-stamps are from free authorities and every page says so To start A contract and a call per document A database to run or a cloud account A free key with no form, one HTTPS call, drop-in recorders, or no integration at all If the provider stops The token still checks against the authority’s certificate Depends on what the vendor lets you export Your proofs still check, offline, without us ( continuity plan ) How it works You keep writing records as you do today The advice note, the decision, the log line stays in your system, in your format. Nothing about your client is sent anywhere. Your system sends a fingerprint One HTTPS call with the record’s salted SHA-256 and a short tag, or a drop-in recorder that does it for you in Python or Node. No SDK is required. Integration guide We seal it where nobody can edit it The database itself refuses any change or deletion, for every role, including ours. Heads are signed, time-stamped by two independent authorities (not qualified under eIDAS) and anchored in Bitcoin. Anyone checks it, any time, without us Hand over the record and its proof. Free verifiers in JavaScript and Python check them offline. Watch a changed record get caught Nothing to lose by trying it Free sandbox, today A key in under a minute, with no form and no sales call. Every endpoint, every proof and both verifiers work against it. No lock-in The format is published, the verifiers are open source and the recorders are Apache-2.0. Take every entry, with its proofs, in one call, whenever you like. The specification Your evidence outlives us RiskRouter is run by one person today. If it stopped tomorrow, every export you hold would still verify. What you keep and how to leave are written down in the continuity and exit plan . Or run it yourself The same service runs in your own cloud account, against your own database, calling nothing of ours. Run it yourself Built for the day it is tested Evidence is only worth something if it holds up when things go wrong, so most of the engineering went there. A failed write is reported as failed If a record cannot be sealed, the answer says so. It never claims a record that does not exist, because that is precisely what an audit looks for. Nothing is lost in an outage The recorders write each record to your own journal first and send it when the log answers again. A test cuts the API mid-run and checks every record arrives. Backed up daily, restored daily Encrypted backups are kept 90 days, and a restore drill rebuilds the chain from them every day. Status and uptime The log limits its own write rate Because nothing can be deleted, a flood would be permanent. The database caps how fast it accepts entries, even if the API in front of it is bypassed. One log, fourteen laws The same log takes the fingerprint of any regulated decision. For each law, a record template says which fields to keep and which article each one answers: IDD, MiFID II, the AI Act, the GDPR, the Consumer Credit Directive, the Anti-Money Laundering Regulation, DORA, NIS2, the Cyber Resilience Act, the VAT Directive, the public procurement directive, the Clinical Trials Regulation and the two medical device regulations. The law names no product, and sealing a record does not make anyone compliant; it makes the record you keep provable. What each law asks you to record Where this stands today RiskRouter runs as a validation build. The evidence log, its proofs and the verifiers are live and usable today with a free sandbox key. A quote endpoint in the developer documentation shows a decision being priced and recorded in one call; it prices a simulated sample matrix, places no cover, involves no carrier and takes no payment. Production pricing for the evidence log is not published yet. The regulatory position sets out what is and is not in place. Get a free sandbox key Talk about a pilot Read the security answers ## Pricing example: a decision priced and sealed in one call URL: https://riskrouter.eu/console Pricing example a worked example of sealing a decision as it is made; sample rates, no product, no cover Monthly premium 0.00 a month Base 0.00 Options 0.00 Annual 0.00 Run the full route Prices come from the server. This page never calculates a premium and never touches the database. Figures are indicative. No cover is in force. Response Waiting for the first quote. Payload Copy Trace ## Integrate RiskRouter URL: https://riskrouter.eu/integrate Integrate There is no SDK to install and nothing to license. You hash a record you keep, send us only the digest, and get back an entry and a proof that anyone can check without us. Everything below runs against the live validation build right now. Quickstart Get a sandbox key Check your setup Keys Failure cases Limits Integration kit Recorders AI agents Pricing example Run it yourself What you bring Quickstart Record a fingerprint and prove it, in three steps You keep the record. We receive its salted SHA-256 and a short kind , never the record, and answer with an entry in the log. The proof that the entry is in the log, under a head we signed, is checked without us. Recording needs a key: take a sandbox key below (one click, no account), then: export RISKROUTER_API_KEY=rr_sandbox_... # 1. Fingerprint a record you keep. Salt it, or anyone who can guess the record can confirm it. SALT=$(openssl rand -hex 16) DIGEST=$( (printf '%s' "$SALT"; cat decision.json) | sha256sum | cut -d' ' -f1) # macOS: shasum -a 256 # 2. Record the digest. Only "recorded": true means there is an entry. curl -s https://api.riskrouter.eu/api/v2/evidence \ -H "Authorization: Bearer $RISKROUTER_API_KEY" -H 'Content-Type: application/json' \ -d "{\"record_digest\":\"$DIGEST\",\"kind\":\"ai.decision\"}" | tee entry.json # 3. Fetch the entry's proof and check it offline, with the public verifier. LEAF=$(python3 -c 'import json; print(json.load(open("entry.json"))["entry"]["leaf_index"])') curl -s "https://api.riskrouter.eu/api/v2/evidence/proof?leaf_index=$LEAF" \ -H "Authorization: Bearer $RISKROUTER_API_KEY" > proof.json git clone https://github.com/aniljangrabe-dev/riskrouter-verify python3 riskrouter-verify/tools/python/riskrouter_verify.py proof.json --key riskrouter-verify/anchors/ The verifier prints two lines: that the entry is in the tree under the root the proof names, and that we signed that head. It uses no network and no key of ours beyond the public one, in a repository that shares no code with this service. Keep decision.json , the salt and entry.json together: without the record, nobody, including us, can show what it said. From a Node backend import { createHash, randomBytes } from 'node:crypto'; import { writeFileSync } from 'node:fs'; // Your record stays with you. Only its salted digest is sent. const record = JSON.stringify({ case: 'dec-2292', outcome: 'refer', model: 'credit-v7' }); const salt = randomBytes(16).toString('hex'); const record_digest = createHash('sha256').update(salt + record).digest('hex'); const res = await fetch('https://api.riskrouter.eu/api/v2/evidence', { method: 'POST', headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${process.env.RISKROUTER_API_KEY}` }, body: JSON.stringify({ record_digest, kind: 'ai.decision' }) }); const out = await res.json(); // 201 with recorded: true is the only success. A refused write is a 503 with recorded: false. if (res.status !== 201 || out.recorded !== true) throw new Error(out.error?.code ?? `HTTP ${res.status}`); writeFileSync(`evidence-${out.entry.leaf_index}.json`, JSON.stringify({ record, salt, entry: out.entry })); From Python import hashlib, json, os, secrets, urllib.request # Your record stays with you. Only its salted digest is sent. record = json.dumps({"case": "dec-2292", "outcome": "refer", "model": "credit-v7"}) salt = secrets.token_hex(16) digest = hashlib.sha256((salt + record).encode()).hexdigest() req = urllib.request.Request( "https://api.riskrouter.eu/api/v2/evidence", data=json.dumps({"record_digest": digest, "kind": "ai.decision"}).encode(), headers={"Content-Type": "application/json", "Authorization": "Bearer " + os.environ["RISKROUTER_API_KEY"]}, ) with urllib.request.urlopen(req, timeout=10) as r: # raises on 401, 429 and 503 out = json.load(r) assert out["recorded"] is True with open("evidence-%d.json" % out["entry"]["leaf_index"], "w") as f: json.dump({"record": record, "salt": salt, "entry": out["entry"]}, f) For a record someone else must be able to re-hash byte for byte, use the canonical form the regime packs define: they also say which fields to record for IDD, MiFID II, the AI Act, the GDPR, DORA and others. The recorders do the canonical form, the salt and the journal for you. Nothing to integrate yet? Seal a file in the browser. Get a sandbox key Record and verify in five minutes, without writing to anyone A sandbox key records entries in the real log, exactly like a key we issue by hand. You get it on this page, once. We ask for nothing about you and store nothing about you: no account, no email, no name. Only a SHA-256 digest of the key is kept. I understand that entries I record with this key are written to an append-only log and can never be deleted , by me or by RiskRouter. Leave this field empty Get a sandbox key Sandbox only 50 recorded entries per key per day limits are enforced by the database, not by this page. A production key is issued by a person, after a conversation. Your sandbox key — copy it now Copy key It is shown once and cannot be recovered, because we only keep its digest. Lost it? Take another. Want a person to look at your use case? Write to us and quote the prefix — it is how we find your sandbox without you sending the key. Check your setup Is your key recording, and does your evidence check out? Paste a key and the number of an entry you recorded with it (the leaf_index the log answered with). This page fetches that entry’s proof and checks it here: that the key is recognised, that the entry is yours, that it sits in the log under the head the proof carries, and that we signed that head. If the key has also recorded quotes through the pricing example , their export is checked too. The key is sent only to api.riskrouter.eu , is not stored, and is gone when you leave the page. Your key Entry number Check my setup Keys Reading is open. Recording is attributed. Every entry is a leaf of an RFC 6962 Merkle tree, and every head of that tree is signed and can be saved by anyone. That is what makes an entry worth something to an auditor, and it is also why the write path cannot be open: an entry is permanent, and the distributor whose key recorded it is named inside the leaf, so who recorded it cannot be rewritten later. You send You get Log No key Signed heads, checkpoints, consistency proofs and public keys: everything needed to check a proof Nothing written. A write answers 401 NOT_AUTHENTICATED A valid key Your entry, and the proof of any entry you recorded Recorded against your distributor, inside the leaf An invalid, revoked or suspended key HTTP 401 Nothing written A bad key is refused outright and never quietly treated as no key. The alternative looks friendlier and is far worse: a firm whose key was mistyped or revoked would keep sending digests for months, believing it was building evidence, and would find out during the audit that nothing had been written. Loud now beats silent then. We store the SHA-256 of your key, never the key. That means we cannot show it to you again, cannot email it to you, and cannot recover it if you lose it; we can only revoke it and issue another. It also means a breach of our database does not hand anyone the ability to record in your name. Getting your evidence back out Each entry has its own proof, fetched with the same key. Fetch it when you record, and keep it with the record and the salt: the three together are the evidence, and they verify without us. Now and then, save a signed head as well; it needs no key. Every later head must be consistent with it, and GET /api/v2/evidence/consistency proves that it is, so a log rewritten after you saved the head cannot pass. curl -s "https://api.riskrouter.eu/api/v2/evidence/proof?leaf_index=$LEAF" \ -H "Authorization: Bearer $RISKROUTER_API_KEY" > "proof-$LEAF.json" curl -s https://api.riskrouter.eu/api/v2/evidence/head > "head-$(date +%F).json" Both verifiers, standard-library Python and Node, are in github.com/aniljangrabe-dev/riskrouter-verify , a separate, public repository that shares no code with this one. Keep them with your proofs. A proof you saved last year still verifies after we have stopped answering the phone, which is the only honest answer to what happens to my evidence if you go under . It is also why there is no SDK and nothing to license: a product that is hard to leave is a product that has to be, and we would rather not need that. Failure cases Handle these six before you ship Each one is cheap to handle now and expensive to discover in an audit. What happens What to do A 503 with recorded: false The log refused the write: an outage, or a fault on our side. Nothing was recorded. Keep the record and its digest on your side, send it again later, and never mark it recorded until a 201 says so. The recorders keep a journal and do this for you. A 429 EVIDENCE_CEILING for your key or for the whole log, or too many requests from your address. Nothing was recorded. Back off and send it again. A sandbox key records at most 50 entries a day; see Limits . A 401 where a 201 used to come back Your key was revoked, or your distributor was suspended. Treat it as an outage and page someone. Do not drop the key and carry on: from that moment nothing you send is evidence. A timeout, and you send it again The first attempt may have been recorded before the connection dropped. Number your records in a completeness chain ( chain_tag and chain_seq beside the digest). A position already taken is refused with 409 CHAIN_OUT_OF_TURN and the next free position, so a retry cannot record twice and tells you the first attempt landed. Without a chain, a retry can leave the same digest twice: harmless, but it says you recorded it twice. The record or the salt is lost The digest alone shows nothing to anyone. Store the record, the salt and the entry together, before you send the digest, and back them up like any record you must keep. We cannot give any of them back. A digest of an unsalted record A short or predictable record can be guessed and the guess confirmed from its digest. Always hash a random salt of at least 16 bytes in front of the record, and keep it with the record. Limits What the log refuses Limit Value On breach Request body 16 KB 413 PAYLOAD_TOO_LARGE One batch ( /api/v2/evidence/batch ) 500 entries, 128 KB Refused whole: nothing in it is written Entries, whole log 600 / min 429 EVIDENCE_CEILING , nothing recorded Entries, one self-serve sandbox key 50 / day 429 EVIDENCE_CEILING , nothing recorded Entries, all self-serve sandbox keys together 30 / min 429 EVIDENCE_CEILING , nothing recorded Requests from one address 60 / min 429 The entry ceilings are enforced by the database, not by this page or the Worker. The log refuses deletion, so a flood would be permanent damage, and a cap that no caller can talk its way around is worth more than a generous one that can be. The sandbox ceilings are lower so that a key we issue by hand can always record. Integration kit Working code in the language your product is written in Each example records a salted fingerprint of a demands-and-needs note and proves that entry, and runs the pricing example too: it prices a quote, records it with an idempotency key, retries without recording twice and exports the evidence. Standard library only, no SDK to install, and every one is run on every change against the real service, with its output checked by two independent verifiers. Python : riskrouter.py , signer.py (sign your own claims), demo.py PHP : RiskRouter.php , demo.php C# (.NET 8) : RiskRouter.cs , Program.cs , RiskRouterDemo.csproj Java : RiskRouterDemo.java Reference app : app.py , a one-file broker desk that keeps the customer’s note on your side and produces an evidence pack an inspector can check offline with check_pack.py . Verifier : riskrouter_verify.py , standard-library Python, independent of everything else here. How to use the kit: README . A verification page in your own brand : verifier-page.mjs writes one HTML file, with your name and colour, that checks a receipt or a disclosure in your customer’s browser and whose security policy forbids it any network request. example-brand.json Drop-in recorders Evidence from where your decisions are already logged If your system already logs its decisions, through Python’s logging , through OpenTelemetry spans, or through a function that returns a decision, a recorder turns that into evidence without an integration project. It canonicalises each record, salts it and keeps it, with its salt, in a journal on your side; only the digest and the kind are sent, in batches, to this log, to your own instance, or to none. A record is written to the journal before its digest is sent, so an outage never loses one, and it is marked recorded only when the log says so. No dependencies, and never a condition of using the API. # Python: a log line with extra={"evidence": record} becomes evidence log.addHandler(EvidenceHandler(Recorder(api_key=KEY, journal="evidence.jsonl"))) log.info("referred", extra={"evidence": {"kind": "ai.decision", "record_id": "dec-2292", "outcome": "refer"}}) // Node: OpenTelemetry spans with evidence.kind become evidence provider.addSpanProcessor(new EvidenceSpanProcessor(createRecorder({ apiKey: KEY, journal: './evidence.jsonl' }))); Python (3.8+, standard library): pip install riskrouter-recorder , or the files: riskrouter_recorder , with a logging handler, a decorator and an OpenTelemetry span processor, and agents.py . README Node (20+, no dependencies): npm install riskrouter-recorder , or the files: recorder.mjs , with withEvidence and an OpenTelemetry span processor; agents.mjs and mcp-server.mjs . README AI agents. The same packages carry adapters for the frameworks agents are built with: a callback handler for LangChain and LangGraph (Python and JavaScript) and a tracing processor for the OpenAI Agents SDK (both languages). Each agent run becomes one record, with its tool calls, model calls, handoffs, guardrails and graph nodes as steps; each step keeps a SHA-256 of the text it handled, not the text. An MCP server gives any MCP client a record_decision tool; what it records is the agent’s own statement, and says so. # LangGraph: every run of the graph becomes one record graph.invoke(state, config={"callbacks": [EvidenceCallbackHandler(recorder, fields={"system_ref": "claims-agent"})]}) // OpenAI Agents SDK (JavaScript) addTraceProcessor(new EvidenceTracingProcessor(recorder, { fields: { system_ref: 'claims-agent' } })); bundle(record_digest) gives you the record, its salt and the proof, which a supervisor checks offline with either verifier. Both recorders are tested on every change against the real engine and the real OpenTelemetry, LangChain, OpenAI Agents and MCP SDKs. They are licensed Apache-2.0, like the verifiers, the conformance suite and the integration kit above. Both are published as riskrouter-recorder , on PyPI and npm ; the files above are the same code, if you would rather copy them. Nothing left out. Give a recorder a chain secret and every record goes in a numbered completeness chain : the recorder adds the chain fields itself and carries on from its journal after a restart, and completeness() returns the log’s signed list of the chain with every record in it, ready for an auditor to check that none is missing. recorder = Recorder(chain_secret=os.environ["RISKROUTER_CHAIN_SECRET"], chain="kyc-desk") At the OpenTelemetry Collector. otlp-sealer.mjs takes OTLP/HTTP JSON from a Collector like any other backend, and turns spans and log records that carry evidence.kind into records, with no library inside the application. A batch is answered only once it is in the journal, so the Collector retries anything not kept. RISKROUTER_API_KEY=... node otlp-sealer.mjs --port 4319 --journal /var/lib/evidence/journal.jsonl --chain otel Files kept elsewhere. A record can name a file it does not contain, such as a call recording, an invoice or a release: digest_file(path) in Python, digestFile(path) in Node, streams its SHA-256 and the file stays where it is. Rows in your database. dbseal.mjs seals the rows of a PostgreSQL table as they are written: it prints the SQL for a queue and a trigger, which you read and apply, then seals each inserted, updated or deleted row by the SHA-256 of its text. The row stays in your database, and a row is marked sealed only once the log has recorded it. verify shows a row changed with the trigger off. node dbseal.mjs install --table public.suitability_reports --key id | psql "$DATABASE_URL" RISKROUTER_API_KEY=... node dbseal.mjs run --db "$DATABASE_URL" --chain Files as they land. folderseal.mjs watches a folder (a shared drive, a synced SharePoint or OneDrive library) and seals each new or changed file, the way /seal does, with nobody remembering to. Each receipt checks at /check . Nothing is uploaded. RISKROUTER_API_KEY=... node folderseal.mjs --folder ./advice-files --chain The inspector’s file. inspector.mjs puts everything a journal sealed in a period into one folder a supervisor checks without us: a receipt per record, each chain shown complete, the log’s public key, a page listing every record, and a sealed manifest. RISKROUTER_API_KEY=... node inspector.mjs --journal evidence-journal.jsonl --from 2026-10-01 --to 2026-10-31 --out ./inspector-2026-10 Security logs. logseal.mjs seals any log stream ( journalctl , tail -F , a syslog relay) segment by segment, so after an incident you can show that no sealed segment was changed or removed. The segments stay in your own archive. Sealed security logs AI agents Connect an AI agent over MCP The evidence log speaks the Model Context Protocol, so an AI assistant or agent framework can read the log, fetch proofs and record digests without anyone writing an integration. Add this server to any MCP client: https://api.riskrouter.eu/mcp { "mcpServers": { "riskrouter": { "type": "http", "url": "https://api.riskrouter.eu/mcp", "headers": { "Authorization": "Bearer " } } } } Reading needs no key. record_evidence takes the SHA-256 of a record you keep and a short kind, never the record, and needs your key; a sandbox key takes a minute. Each tool is the same route as in the API reference , with the same checks. There is no pricing tool. To record from inside the agent’s own process instead, with the record kept in your journal, use the local MCP server and the framework adapters above. In Claude, add it from the connector directory: RiskRouter evidence log . Every client’s setup in one place: AI agents and assistants . Worked example The pricing engine: a deterministic engine whose every answer can be recorded RiskRouter began as a pricing engine for micro-insurance. It stays as a worked example for developers, not as an offer: a deterministic engine whose quotes can be recorded in a hash-chained ledger, the older cousin of the evidence log above. Its rating matrix is in the API reference , and the console drives it. Nothing above depends on it. Price a configuration No key: a real price, and nothing written. With a valid key, the quote is also recorded against your distributor, and attribution.recorded says whether it landed. An invalid key is refused with 401, never priced as anonymous. curl -X POST https://api.riskrouter.eu/api/v1/quote \ -H 'Content-Type: application/json' \ -H "Authorization: Bearer $RISKROUTER_API_KEY" \ -H "Idempotency-Key: order-4711" \ -d '{ "active_vertical": "mobility", "selected_components": { "commuter_roadside": true } }' // Two different questions. Ask both. const priced = quote.ok === true; // is this number real const recorded = quote.attribution.recorded === true; // is it in the ledger Ask the engine what it charges Do not hardcode the price list. GET /api/v1/verticals publishes every vertical, its mandatory base layer and its optional modules, with the prices the engine will apply. Build an interface from that response and a price change never desynchronises it from the quote. What to check on every quote persistence.persisted : never say a quote is on file unless it is literally true . persistence.attempted tells no key apart from a write the ledger refused. ignored_components : anything in it is a component the engine does not price, so a deployment mismatch. Alert on it. An Idempotency-Key on every recorded quote, reused on every retry of that quote: the same key records at most one entry, and a retry gets it back with persistence.replayed: true . Never show a number your own code calculated, and never fall back to one when the engine does not answer. Limits: a 16 KB body, a 5-second ledger write and 120 ledger writes a minute across all callers. Past either of the last two the quote still returns, marked not persisted. Export the quote ledger curl -s -H "Authorization: Bearer $RISKROUTER_API_KEY" \ https://api.riskrouter.eu/api/v1/ledger/export > "ledger-$(date +%F).json" node riskrouter-verify/tools/verify-attestation.mjs ledger-2026-09-20.json # we signed this head node riskrouter-verify/tools/verify-ledger.mjs ledger-2026-09-20.json # your entries produce it You get your entries in full, the digest skeleton of the whole chain they sit inside, and the signed attestation for the head. /verify checks the same export in the browser. No lock-in Run it yourself The service is one Worker and one PostgreSQL schema, with no proprietary runtime. A deployment kit runs that same code in your own account: on your own Cloudflare and Supabase, or anywhere containers run. Your instance signs its heads with its own key, holds its own data and calls nothing of ours: no licence server, no telemetry. The verifiers, the recorders and the conformance suite work against it exactly as they work against ours, and a sandbox key from your instance carries your instance’s address in its next steps. The service’s source is in a private repository today (the verifiers, the recorders and the kits are what is published), so ask us for it ; if running it without us is a condition of your pilot, agree how you would get it before the pilot starts. See the continuity and exit plan . We would rather you could leave than have you locked in. An integration you can walk away from is one you can commit to. Before a pilot What you bring RiskRouter is evidence infrastructure. An entry shows that a record with that digest existed no later than the entry’s time and has not changed since. It does not show that the record is true or complete, and it cannot stand in for the things a supervisor expects you to hold. The records themselves, and their salts, kept and backed up on your side for as long as your rules require Your own decision about what goes in a record. The regime packs map fields to articles as orientation, not as legal advice Your own authorisation where your activity needs one: an FSMA registration, or its equivalent in your member state Your own terms with your customers, and your own data-protection analysis of what you record We supply the log, the proofs and the verifiers. The regulatory position explains where the boundary sits and why it is drawn there. Start a pilot conversation Full API reference ## API reference · RiskRouter URL: https://riskrouter.eu/docs API reference Every endpoint, and no personal data in anything the API stores. Pricing needs no authentication; recording a quote in the audit ledger needs a key, and you can get a sandbox key right now. The server prices every quote and discards any price a client sends. Base URL https://api.riskrouter.eu Machine-readable: openapi.json (OpenAPI 3.1). Point your client generator or coding assistant at it; a test in the repository keeps it identical to the routes the Worker actually answers. Or import riskrouter.postman_collection.json into Postman, Insomnia, Bruno or Hoppscotch. It is generated from the same spec on every build. Its first request takes a sandbox key and stores it, so everything after it works without copying anything; the recording request sends a fresh Idempotency-Key each time. Or skip the tooling and try it on this page , or start from working code in Python, PHP, C# or Java in the integration kit . Browsers may call this API only from the console origin. Server-to-server callers are unaffected, because the browser enforces that rule rather than the API. Try it here Real requests to api.riskrouter.eu , sent from this page. Pricing needs no key. To record, paste a sandbox key ( take one in a few seconds ); it stays in this tab’s memory, is sent only to our API, and is gone when you close the tab. Anything you record is permanent, so record test configurations, never personal data. Request POST /api/v1/quote, no key (priced, not recorded) POST /api/v1/quote, with your key (recorded, idempotent) GET /api/v1/verticals GET /api/v1/ledger/attestation GET /api/v1/ledger/export, with your key Sandbox key Body { "active_vertical": "mobility", "selected_components": { "commuter_roadside": true } } Send POST /api/v1/quote Prices a configuration, and appends an audit record when a distributor key is presented. The only endpoint that writes to the ledger. Authentication Optional, and what it changes is whether the quote is recorded rather than whether it is priced. The ledger is tamper-evident, so an entry cannot be cleaned up later without breaking every digest after it. That is why writes are attributed and reads of the price are not. Header Response Ledger Absent 200 with a real price Nothing written. attribution.mode is ANONYMOUS Authorization: Bearer 200 with a real price Recorded against that distributor, inside the hash chain A key that does not resolve 401 INVALID_API_KEY Nothing written An unresolvable key is refused rather than downgraded to an anonymous quote. Downgrading would let a distributor with a mistyped or revoked key collect 200s for months and discover at audit that nothing had been recorded. We store only the SHA-256 of a key, so a lost key can be revoked and replaced but never recovered. Retries: Idempotency-Key Optional, and worth sending on every recorded quote. The same key from the same distributor records at most one ledger entry, ever; a retry with it gets that entry back with persistence.replayed: true and an Idempotent-Replayed: true header, and nothing new is written. The check and the write happen in one database transaction, so two retries racing each other still produce one entry. Use 1 255 visible ASCII characters with no spaces; your own order or session reference is ideal. Only its SHA-256 is stored, never the key, so nothing you put in it can end up in a table that is never deleted; a random UUID is still the better habit. Reusing a key for a different quote is refused with 409 IDEMPOTENCY_KEY_REUSED and records nothing. Different means different in what would be recorded — vertical, components the engine actually priced, premium — so a retry that adds a component the engine ignores is still the same request. On an anonymous quote the header is ignored, because nothing is written. Request Send active_vertical as one of the four keys below. selected_components is optional; leave it out to price the base layer alone. Only an explicit true enables a component, so the string "false" selects nothing. curl -X POST https://api.riskrouter.eu/api/v1/quote \ -H 'Content-Type: application/json' \ -d '{ "active_vertical": "urban_rental", "selected_components": { "deposit_liquidity_swap": true, "smart_home_iot": true } }' Response selected_components comes back normalised to what was actually sold. The mandatory base layer is always present and always true. Anything the engine did not recognise appears in ignored_components , so a client that has drifted out of step shows up in testing. { "ok": true, "quote": { "quote_reference": "…", "active_vertical": "urban_rental", "currency": "EUR", "billing_interval": "MONTHLY", "selected_components": { "tenant_fire_water": true, … }, "ignored_components": [], "line_items": [ … ], "totals": { "base_layer_premium": "3.50", "add_on_premium": "6.00", "total_monthly_premium": "9.50", "indicative_annual_premium": "114.00", "active_add_on_count": 2 } }, "compliance": { "status": "MOCK_ROUTING_VERIFIED", "binding_state": "NOT_BOUND. Indicative simulation only, …" }, "persistence": { "attempted": true, "persisted": true, "record_id": "…", "latency_ms": 41 }, "attribution": { "recorded": true, "mode": "ATTRIBUTED", "distributor": "…", "environment": "sandbox" } } Reading persistence A 200 does not prove an audit record exists. The write is reported separately, and a failed write degrades the response instead of failing it, because a ledger fault should not take pricing down with it. Treat a record as stored only when persistence.persisted is literally true . persistence.attempted separates the two reasons it can be false: no key was sent, or the write was tried and did not confirm. The first is a configuration choice, the second is an incident. POST /api/v1/sandbox/keys Issues a sandbox key on the spot, with no account and no one to write to first. The key records quotes in the ledger exactly like one we issue by hand, so the whole evaluation — price, record, export, verify — can be done in a few minutes. Or use the form on the integration page , which calls this for you. curl -s -X POST https://api.riskrouter.eu/api/v1/sandbox/keys \ -H "Content-Type: application/json" \ -d '{"acknowledge_permanent": true}' acknowledge_permanent is required and must be the boolean true . Entries you record are written to an append-only ledger and can never be deleted, and you should not find that out afterwards. Nothing else is read from the body: no name, no email, nothing about you is stored. The response carries the key once, its prefix, and runnable next steps with the key already in them. Only a SHA-256 digest of the key is kept, so a lost key cannot be recovered — ask for another. Limits are enforced by the database itself: 50 recorded entries per key per day, and a shared allowance of new keys per hour. Sandbox only; a production key is issued by a person. GET /api/v1/verticals Publishes the rating matrix, so a client renders exactly what the engine prices instead of carrying a copy that can drift. The console loads this on startup. GET /api/v1/ledger/attestation The entry count, the head hash, and the canonical form the digest is computed over. Record the head with a date and any later history that does not reproduce it has been altered. Nothing about any customer is exposed: the attestation commits to the ledger without disclosing it. GET /api/v1/ledger/export Your own entries in full, plus (chain_index, row_hash, prev_hash) for every entry in the ledger, plus the signed attestation. Requires your API key: an export is scoped to one distributor, and an anonymous caller has nothing to export. Together those three pieces let you rebuild each of your digests from its own content and follow the chain to the head we signed — with no credential of ours and no call to us at verification time. The skeleton carries digests only, never another distributor’s content. curl -s -H "Authorization: Bearer $RISKROUTER_API_KEY" \ https://api.riskrouter.eu/api/v1/ledger/export > export.json node tools/verify-attestation.mjs export.json # we signed this head node tools/verify-ledger.mjs export.json # your entries produce it A ledger with more than 25 000 entries is refused with LEDGER_TOO_LARGE rather than truncated. A partial skeleton cannot reach the head, so it would verify as broken and send you hunting for tampering that never happened. Rate limited twice, because two different people abuse it. By address before your key is resolved, so a stranger cannot spend a database round trip per request just by asking; then six a minute per distributor, keyed to you rather than to your address. An export is a periodic record, not a polling endpoint — take one a month and keep it. One export proves consistency. Several, saved months apart, prove history — see verify for why that distinction is the whole point. GET /api/v1/ledger/pubkey The public half of the attestation signing key, and its key id. Useful for a quick check, but not the copy to trust: verifying a signature against a key we hand you at the moment you verify proves nothing if we are the thing in question. The copy that counts is anchors/signing-key.json in the public repository, which stays there across rotations so an attestation saved years ago is still checkable. POST /api/v1/contact Serves this site’s own contact form. It is listed because an undocumented public write endpoint is a thing an auditor is right to ask about, not because it is part of the integration surface — there is no reason to call it from your own code. It writes to contact_enquiries , which unlike the ledger is deletable by design: it holds personal data and must stay erasable under GDPR Article 17. The data protection page sets out what is kept and for how long. POST /api/v1/landings Counts one arrival from a tagged campaign link, for this site’s own measurement; there is no reason to call it from an integration. The body is {"src": "linkedin-sept", "path": "/integrate"} and nothing else is read or stored: no IP address, no identifier, no browser details. The result is a daily count per campaign and page, described on the data protection page . Anyone can inflate a count, so the numbers are marketing signals, never evidence. Answers 204. v2 Evidence log POST /api/v2/evidence Record that any regulated decision existed, without sending it: a suitability assessment, a credit decision, an AI system’s log line, a quote. You hash your own record and send only the digest; we never see the record, so it may contain whatever your own obligations require. Needs your key. Entries are leaves of an RFC 6962 Merkle tree, so one entry is proven against a signed head with a handful of hashes, and nothing about any other entry is revealed. The format is in the open specification . # Salt the record, or anyone who can guess it can confirm it from the digest. SALT=$(openssl rand -hex 16) DIGEST=$( (printf '%s' "$SALT"; cat decision.json) | sha256sum | cut -d' ' -f1) curl -s -X POST https://api.riskrouter.eu/api/v2/evidence \ -H "Authorization: Bearer $RISKROUTER_API_KEY" -H "Content-Type: application/json" \ -d "{\"record_digest\":\"$DIGEST\",\"kind\":\"ai.decision\"}" Two fields only: record_digest (lowercase hex SHA-256) and kind (a tag you choose: a z, 0 9, dots and hyphens, at most 40, such as mifid.suitability ). Anything else is refused, never coerced. The answer is 201 with the entry ( leaf_index , created_at , leaf_hash ) and recorded: true ; a refused write answers 503 with recorded: false , never a success. Keep your record, your salt and the entry: without your record nobody, including us, can show what it said. Not sure what to put in the record? The regime packs say, for IDD, MiFID II, the AI Act, GDPR Article 22, DORA and NIS2, which fields to record and which article each one speaks to, with a canonical form so someone else can re-hash your record exactly. Optionally, sign the claim yourself. Send signer_key_id , claimed_at (your own timestamp, YYYY-MM-DDTHH:MM:SS.ffffffZ ) and client_signature (ECDSA P-256 over riskrouter-evidence-claim|v3|kind|record_digest|signer_key_id|claimed_at , raw r || s in base64) with the two fields above, all three or none. The signature is verified against a key you registered before anything is written, and then sits inside the leaf: a v3 leaf, whose proof shows that you asserted the decision, not only that we sealed it. A signature that does not verify records nothing. The specification has the exact forms; the Python kit has a standard-library signer. POST /api/v2/evidence/keys Register the public half of the key you sign claims with, once: public_key as a P-256 JWK ( {"kty":"EC","crv":"P-256","x":"…","y":"…"} ). Needs your key. Answers 201 with the key_id (the first 16 hex characters of SHA-256 over the JSON array [x, y] ) or 200 if that key was already yours. A key id belongs to one distributor forever, so another distributor’s key answers 409 . Keys are never removed: leaves signed before you rotate stay verifiable. Publish the same key somewhere you control, so a verifier need not take our word for whose it is. GET /api/v2/evidence/keys ?key_id=… . The registered public key under that id, for anyone: a public key is public, and a verifier of a v3 leaf needs it. Answers the key and when it was registered, never whose it is. POST /api/v2/evidence/batch Many digests in one request: entries , an array of 1 to 500 objects each with record_digest and kind , recorded in the order sent, in one transaction. The same checks apply to each entry as to a single record, and the same ceilings hold per row: if any entry is refused, nothing in the batch is written and the answer says so. The body may be up to 128 KB. Answers 201 with every entry and its leaf index. GET /api/v2/evidence/head The signed tree head: tree_size , root_hash and timestamp , signed over riskrouter-evidence-head|v2|tree_size|root_hash|timestamp with the same key as the v1 attestation. Unsigned heads say so. Save one; every later head must be consistent with it. GET /api/v2/evidence/proof ?leaf_index=N , optionally &tree_size=M . The inclusion proof for one of your entries (needs your key; another distributor’s entry answers exactly like a missing one): the entry, the audit path, and the signed head when tree_size is the current size. To prove against a head you saved earlier, pass its tree_size and check against its root. Take the size and the root from the same signed head. For a signed (v3) leaf the proof also carries signer_public_key , so the client signature can be checked without another request. GET /api/v2/evidence/consistency ?first=M , optionally &second=N . Proof that the tree you saved at size M is a prefix of the tree at size N : the check that catches a rewritten history. Public, because it reveals nothing but hashes. Check it against the root in the head you saved, never a root we supply. This is what an independent witness runs. GET /api/v2/evidence/chain ?chain_tag=… , optionally &from= and &to= . What the log holds in one completeness chain as of the current signed head: its length and, for each position, the leaf index, the time it was sealed and the record digest, signed over riskrouter-evidence-chain|v1|chain_tag|length|tree_size|root_hash|timestamp|from|to|links_digest with the same key as the head. Up to 1,000 positions a page; next names the following page. Public, like a head: the tag is derived from a secret you keep, so only someone you give it to can ask, and the answer names no distributor and holds no record. To put a record in a chain, send chain_tag (64 hex characters) and chain_seq (1, 2, 3, …) with its digest to POST /api/v2/evidence or in each batch entry, and put the same two inside the record with chain_prev , the digest of the record before it. A position out of turn answers 409 with the position expected; a tag another distributor uses answers 409 ; nothing is written either way. A record a firm then leaves out of what it shows is a gap the statement names. The specification has the forms; tools/chains.mjs checks a chain, and a spot check of it, offline. POST /api/v2/evidence/statement Register a SCITT Signed Statement (body application/cose , your key). It must be a hash envelope: a COSE_Sign1 signed ES256 with a key you registered (its id as kid ), with CWT claims iss and sub and a payload that is the 32-byte SHA-256 of your artifact, so the artifact never comes here. The log keeps one leaf holding the statement’s digest, and answers with an RFC 9942 COSE Receipt and the Transparent Statement, both base64. The profile is in the specification . GET /api/v2/evidence/receipt ?leaf_index=N , with the key that recorded the entry: a COSE Receipt of inclusion for it at the current tree size ( application/scitt-receipt+cose ), for a statement or any other entry. Answered only when the signing key is bound; otherwise 503 RECEIPT_NOT_SIGNED . GET /tlog/evidence-v2/checkpoint The same tree as /api/v2/evidence/head , as a C2SP checkpoint ( text/plain ): a signed note of three lines, the origin api.riskrouter.eu/tlog/evidence-v2 , the tree size and the root in base64, signed with the log's Ed25519 key. This is the form the public witness network cosigns. Answered only when that key is bound and matches the published one; otherwise 503 CHECKPOINT_NOT_SIGNED , never an unsigned note. Format and key in the specification . GET /tlog/evidence-v2/tile//[.p/] The tree as tiles of 256 hashes, exactly as c2sp.org/tlog-tiles lays them out, so anyone can mirror the whole tree and compute any root or proof without this API. Immutable once served. Only hashes are published: entry bundles ( tile/entries/… ) answer 404 ENTRY_BUNDLES_NOT_PUBLISHED , because every leaf names a distributor and a kind. node tools/checkpoint.mjs mirror fetches every tile and checks it gives the checkpoint's root. POST /mcp The evidence log for AI agents, over the Model Context Protocol (Streamable HTTP, stateless: one JSON-RPC message per request, one JSON answer, no session). Point any MCP client at https://api.riskrouter.eu/mcp . Tools: get_evidence_head , get_inclusion_proof , get_consistency_proof , get_checkpoint , get_ledger_attestation , get_usage , get_chain and record_evidence . Each is the route above it on this page, answered by the same code with the same errors. Reading needs no key; record_evidence takes a digest and a kind, never a record, and needs your key in the connection's Authorization header, or a key the client gets by signing in . There is no pricing tool: the engine prices for licensed distributors' software, not for an agent talking to the public. Connecting a client . GET /.well-known/oauth-protected-resource GET /.well-known/oauth-protected-resource/mcp GET /.well-known/oauth-authorization-server POST /oauth/register GET POST /oauth/authorize POST /oauth/token Sign-in for MCP clients that cannot send a key header, such as the connectors in Claude's directory. A tool that needs a key, called with none, answers 401 with a WWW-Authenticate header that points at the protected-resource metadata; the client then registers (no secret), sends the user to /oauth/authorize , where they tick that entries are permanent, and exchanges the code (PKCE S256 required, valid two minutes) at /oauth/token . The access token is a new self-serve sandbox key, issued exactly as POST /api/v1/sandbox/keys issues one, under the same limits; nothing about the user is asked or stored. Reading never needs it. GET /api/v1/usage How much is recorded, and by whom, as counts: entries in the quote ledger and the evidence log, the last 7 and 28 days, twelve calendar weeks, and the figure that matters, outside : entries recorded with keys that are not ours, self-serve sandbox keys counted separately. Public, because it returns nothing but numbers and week dates; no identifier, no premium, no record. Cached for five minutes at the edge. /usage shows it. GET /api/v1/health Liveness, plus whether the audit ledger binding resolved. binding_configured: false means quotes still price but nothing is recorded. The bare domain answers this too. Errors Every error carries the same envelope: ok: false and an error object with a stable code , a message , and a docs link to its row in this table. Code HTTP Cause UNKNOWN_VERTICAL 400 Missing or unrecognised active_vertical INVALID_COMPONENTS 400 selected_components was an array rather than an object INVALID_PAYLOAD 400 Body was not a JSON object MALFORMED_JSON 400 Body was not valid JSON INVALID_API_KEY 401 Key is unknown, revoked, or its distributor is suspended. Nothing was recorded KEY_CHECK_UNAVAILABLE 503 A key was sent but could not be checked. Refused rather than priced unrecorded NOT_AUTHENTICATED 401 An export was requested with no key. There is nothing to export for nobody in particular LEDGER_TOO_LARGE 413 The ledger is past the single-response export ceiling. Refused rather than truncated into a proof that fails UNSUPPORTED_MEDIA_TYPE 415 Content type was not application/json PAYLOAD_TOO_LARGE 413 Body exceeded 16 KB INVALID_IDEMPOTENCY_KEY 400 Idempotency-Key was empty, too long, or not visible ASCII IDEMPOTENCY_KEY_REUSED 409 That key already recorded a different quote; nothing was written ACKNOWLEDGEMENT_REQUIRED 400 A sandbox key was requested without "acknowledge_permanent": true SANDBOX_CAPACITY_REACHED 429 This hour's allowance of new sandbox keys is used; nothing was issued SANDBOX_NOT_OFFERED 404 A self-hosted instance whose operator has not switched on self-serve sandbox keys; ask the operator for a key. riskrouter.eu always offers them SANDBOX_NOT_ISSUED 503 The key could not be stored, so none was handed over METHOD_NOT_ALLOWED 405 Wrong verb for the route ROUTE_NOT_FOUND 404 No route bound to that path BAD_REQUEST 400 The request URL could not be parsed BODY_READ_FAILED 400 The request body could not be read REQUEST_REFUSED 400 The sign-up form's hidden bot-trap field was filled in; nothing was issued INVALID_ENQUIRY 400 A contact enquiry had fields that need attention; details says which INVALID_LANDING 400 A landing count carried a malformed campaign tag or a path that is not a page on this site RATE_LIMITED 429 Too many requests from one address in a short time. Wait a minute; when a Retry-After header is sent, it says how long ENQUIRY_NOT_STORED 503 The enquiry could not be saved, so it did not reach us. Nothing is reported as sent LANDING_NOT_COUNTED 503 The visit could not be counted; nothing else is affected LEDGER_UNAVAILABLE 502 / 503 The ledger or its attestation could not be read NO_SIGNING_KEY 500 / 503 This Worker publishes no attestation key, or the one bound is unreadable INVALID_EVIDENCE 400 record_digest was not a lowercase SHA-256 hex digest, or kind was not a short tag. Nothing was recorded INVALID_PROOF_REQUEST 400 leaf_index , tree_size , first or second was missing, not a whole number, or outside the tree EVIDENCE_NOT_FOUND 404 No entry at that index was recorded with this key. Another distributor’s entry answers the same EVIDENCE_TREE_TOO_LARGE 413 The evidence log is past the size this route reads in one request. Refused rather than answered with a proof that fails EVIDENCE_CEILING 429 The evidence log is at a ceiling, for this key or overall. Nothing was recorded INVALID_CLIENT_SIGNATURE 400 client_signature does not verify over the claim payload with a key registered to this distributor, or the key id is not one of yours. Nothing was recorded INVALID_SIGNER_KEY 400 public_key was not a P-256 public JWK, or key_id was not a 16-character key id. Nothing was registered CHAIN_OUT_OF_TURN 409 chain_seq is not the next position in that chain; error.details.expected_chain_seq says which is. Nothing was recorded CHAIN_TAKEN 409 That chain_tag is in use by another distributor. Nothing was recorded INVALID_CHAIN_REQUEST 400 chain_tag is not 64 lowercase hex characters, or from or to is not a position in the chain SIGNER_KEY_TAKEN 409 That key is registered to another distributor; a key id belongs to one distributor forever. Nothing was registered SIGNER_KEY_NOT_REGISTERED 503 The key could not be stored, so it is not registered SIGNER_KEY_NOT_FOUND 404 No signing key is registered under that id EVIDENCE_NOT_RECORDED 503 The evidence log did not accept the entry. Answered with recorded: false , never as a success INVALID_STATEMENT 400 The Signed Statement breaks the registration policy (the message names the rule). Nothing was recorded INVALID_STATEMENT_SIGNATURE 400 The statement’s signature does not verify with a key registered to this distributor under its kid. Nothing was recorded RECEIPT_NOT_SIGNED 503 No signing key is bound, so no receipt is issued rather than an unsigned one CHECKPOINT_NOT_SIGNED 503 No checkpoint key is bound, or the bound key is not the published one. No checkpoint is served rather than an unsigned one TILE_NOT_FOUND 404 Not a tile path, or a tile the tree does not have yet ENTRY_BUNDLES_NOT_PUBLISHED 404 Only hash tiles are published; a distributor holds its own entries with their proofs TILES_UNAVAILABLE 503 The stored tree cannot be read for that tile ORCHESTRATION_FAILURE 500 Unhandled fault; no detail is leaked Rating matrix (a worked example) POST /api/v1/quote is a worked example of recording a decision at the moment it is made: the server prices a configuration and, with a key, seals exactly what it priced. The four verticals below are sample data for that example. They are not products, not for sale and not insurance cover, and your own rates or decisions replace them. The live matrix is served at GET /api/v1/verticals , every change is listed with its fingerprint on the changelog , and the demo console drives it. All prices are monthly, in euro. Base layers are mandatory and cannot be switched off. Urban Mobility mobility Component code Cover Premium public_liability Public Liability Base €2.00 battery_hardware_theft Battery / Hardware Theft €1.50 commuter_roadside Commuter Roadside Assistance €1.00 commercial_delivery Commercial Delivery Extension €5.00 carshare_deductible Car-Share Deductible Buy-Down €2.50 Digital Nomad & Freelance digital_nomad Component code Cover Premium cyber_extortion_data Cyber-Extortion & Data Protection Base €1.50 high_value_hardware High-Value Hardware Protection €3.00 international_roaming International Roaming Expansion €2.00 professional_indemnity Freelance Professional Indemnity €4.00 contractual_legal Contractual Dispute Legal Protection €1.50 Urban Rental & Housing urban_rental Component code Cover Premium tenant_fire_water Tenant Fire & Water Liability Base €3.50 deposit_liquidity_swap Security Deposit Liquidity Swap €4.00 landlord_legal_defense Landlord Dispute Legal Defense €1.00 roommate_sublet Roommate / Sublet Liability €1.50 smart_home_iot Smart-Home / IoT Failure Cover €2.00 Micro-Travel & Sports micro_travel Component code Cover Premium multimodal_medical_transit Basic Multi-Modal Medical & Transit Base €2.50 winter_sports_gear Premium Winter Sports & Ski Gear €3.50 parametric_delay Parametric Flight / Train Delay €2.00 adventure_sports Extreme Adventure Sports Addendum €3.00 baggage_tech_transit Lost Baggage / High-Value Tech €1.50 ## For insurance distributors, brokers and MGAs URL: https://riskrouter.eu/for-distributors For insurance brokers, agents and MGAs When a client says you gave the wrong advice, show the note, and prove it is the one you wrote. A claim is refused and the client remembers being told they were covered. What settles it is the demands-and-needs note and your reasons for the recommendation. What weakens it is any doubt that the note in your file is the one you wrote at the time, and not one corrected once the claim came in. Try it in two minutes Send this to your software vendor When it is tested A complaint after a refused claim, taken to the insurance ombudsman; your supervisor’s inspection of your files; a claim against your professional liability cover; a court. What you will be asked for The client’s demands and needs, specified before the contract, and, when you advised, the reasons for what you recommended (IDD, Article 20). What the IDD asks What goes wrong today The note lives in Word, in an email or in your broker software, and any of them can be edited later without a trace an outsider can check. It is your word against the client’s. What changes The note is sealed the day you write it, by your broker software or by hand on this site. In a dispute you hand over the note and its receipt, and the other side checks both themselves, without trusting you or us. Nothing about your client reaches us: only a salted fingerprint. Try it in two minutes Choose a demands-and-needs note or an advice note of your own. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. Most brokers will never integrate anything themselves: their broker software would, once, for everyone on it. Send your vendor this page . Evaluating any system? Start with the twelve-question audit-trail checklist . Where you sit relative to us Money, the customer relationship and the advice note itself stay exactly where they are today. Only a fingerprint reaches us. Your advice the note, made in your own system, as today Your system hashes the note with a salt only you hold and sends us the fingerprint RiskRouter seals the fingerprint into the log and signs the head Your file the note, and proof it has not changed since that day: not by you, not by us, not by anyone with database access The questions your compliance file actually asks Answered directly, not deferred to a call. What data do we send? For an advice note: its salted SHA-256 fingerprint and a short tag such as advice.note . The note, the customer’s name and everything else stay in your system; the fingerprint cannot be turned back into any of them. Does it replace my carriers’ prices? No. Your carriers price their products, as today. What you seal is the record of what you advised. RiskRouter’s own rating engine is optional and prices sample rates ; you do not need it to record evidence. What gets permanently recorded? The fingerprint, its tag, the time and your distributor id, in a log with a signed head. You can prove one record to a supervisor without showing them any other. A priced quote is recorded in full, chained to the entry before it. Where is it hosted? Records are stored in Frankfurt, Germany, inside the EU; requests are processed at Cloudflare’s nearest location, which may be outside the EEA, with nothing stored there. An encrypted daily backup is kept on GitHub for 90 days, outside the EU. The full position is on the privacy page . Can we verify it ourselves? Export your own entries and check them offline with our open-source verifiers, in JavaScript or Python, standard library only, with no call back to us. See how the tamper check works . What if RiskRouter disappears? An export you downloaded last year still verifies after we have stopped answering the phone. That is the actual answer to the actual question, not a promise. How much engineering work is required? If your broker software integrates RiskRouter, none: the note is sealed when you save it. Otherwise, one HTTP call per record and no SDK, with working code in Python, PHP, C#, Java and JavaScript. How do we test it before committing? Take a sandbox key on the integration page , with nobody in the loop. It records in the real log under its own distributor id, so you test the real thing; entries are permanent, and the page asks you to accept that first. If you are an MGA rather than a broker Nothing above changes. The distributor id in the ledger is yours whether you write the recommendation yourself or aggregate it from a panel of agents underneath you — the chain records who was attributed, at the level you issue keys to. What this replaces 01 A spreadsheet nobody can date-stamp A cell that says "reviewed" proves nothing about when, or whether it looked different last month. 02 An argument about which version is real When the note in the file and the one in the CRM disagree, the fingerprint says which one existed on the day. A three-person practice gets the same answer a compliance department would. 03 Trusting the operator's word A database dump is a claim. A chain a holder can verify offline, against a signed head hash, is evidence. Get a sandbox key What your software vendor would do Talk to us ## Trust and verification URL: https://riskrouter.eu/trust For security, compliance and audit review Don’t trust RiskRouter. Verify it. Every claim below is checkable without asking us anything. That is deliberate: a claim you have to take on faith is not evidence, and an audit trail whose only witness is the vendor who produced it is not one either. See tamper detection live Full regulatory position What we send, what we store, where No personal data in a quote A quote payload is a vertical key and boolean component toggles. There is no field for a name, an email, or an identifier — not omitted by policy, absent from the schema. Server-side pricing, always The server looks up the price from its own matrix. A number a client sends is never read back as the answer, so a manipulated frontend cannot produce a cheaper quote. Stored in Frankfurt, processed at the edge Records are stored by Supabase in eu-central-1 , Frankfurt. A request is processed by Cloudflare at the location nearest the caller, which may be outside the EEA; nothing is stored there. We say it this way because EU-only would need a compute restriction we do not have. One more copy exists: a daily backup, encrypted before it leaves with a key GitHub does not hold, is kept on GitHub for 90 days, which is outside the EU. Detail on the privacy page . What the request path can reach, and what it cannot One destination, mechanically enforced — not a policy someone has to remember. The Worker that prices and seals a quote makes exactly one outbound call: to the Supabase URL from its own binding. Nothing else — no analytics, no third-party API, no carrier, no payment rail. A stranger's request cannot cause this Worker to contact anywhere else, and that property is asserted by an automated test on every change, not merely documented. The separate administrative console adds exactly one more destination for its own use: Cloudflare Access's key endpoint, to verify who is signed in. It cannot price a quote, write to the ledger, or reach the public API at all — it does five things only: create a distributor, suspend or reactivate one, issue a credential, revoke one, and mark a contact enquiry as read or answered. The ledger itself 01 Append-only, structurally A database trigger refuses UPDATE and DELETE on the audit table for every role, including the administrator's own. This is not a permission that could be granted back — the trigger fires regardless of who is asking. 02 Chained, not just logged Every entry's hash depends on the entry before it. Changing one historical value breaks every digest computed after it, which is checkable by recomputing them — not by trusting that nobody would. 03 Signed, not just chained We periodically sign the current head hash with an ECDSA key and publish it. A saved, signed attestation is something we cannot later disown — it proves we asserted this history at this time, independent of whether the chain is later rebuilt. Verifying it yourself, offline A hash chain alone proves history was not edited. It cannot prove history was not rebuilt from scratch — a rebuild recomputes every digest consistently and looks perfect. The signed attestation closes that gap: it is a claim we cannot take back. Export your own entries any time you hold a key: curl -s -H "Authorization: Bearer $RISKROUTER_API_KEY" \ https://api.riskrouter.eu/api/v1/ledger/export > ledger-export.json Then run two dependency-free Node scripts against it — standard library only, nothing to install, no call back to us. They live in a separate, public repository that shares no code with the service you are checking, on purpose: git clone https://github.com/aniljangrabe-dev/riskrouter-verify node riskrouter-verify/tools/verify-attestation.mjs ledger-export.json # we signed this head; we cannot deny it node riskrouter-verify/tools/verify-ledger.mjs ledger-export.json # your entries actually produce it Keep the export and the signed attestation somewhere we cannot reach. An export saved today still verifies against everything you hold, whether or not RiskRouter is still answering the phone a year from now. That is the actual answer to what happens if you disappear , not a promise standing in for one. Read the verifier before you trust it; it is short on purpose. If it is useful to you, star it on GitHub so the next compliance officer looking for one can find it. Want to see the chain break in real time instead of reading about it? Try it in your own browser — it lets you tamper with a record and watch the check refuse it. What we do not overclaim Precision matters more here than anywhere else on this site. Not "zero GDPR liability" A quote payload holds no personal data by design, which is a fact about the schema. It is not a legal conclusion about your own obligations, which depend on what you do with the response. Not "regulator-approved" Nobody has approved this. What exists is a record a regulator can independently verify without taking our word for it — a narrower, checkable claim. Not a validation build claiming live rails Every price on this site is a simulation while we validate; the ledger, the evidence log and their proofs are live. See exactly what is and is not in place on the regulatory position page . Go deeper Tamper detection, live Full regulatory position Data protection Ask us anything else ## Regulatory position URL: https://riskrouter.eu/compliance Regulatory position Where this platform sits under EU and Belgian insurance law, what it does with data, and which obligations belong to you rather than to us. Written so that a compliance officer can check it against the system rather than take it on trust. This is not an offer of insurance, and no cover is ever placed. Nothing this platform produces constitutes insurance mediation, distribution or advice within the meaning of Directive (EU) 2016/97. No policy is bound, no carrier is notified, no premium is collected and no payment instrument is touched at any point. What RiskRouter is Current status Frameworks IDD in the design Your obligations Limits of this page Position Infrastructure, not an intermediary RiskRouter supplies software: a pricing engine, a configurator and an audit ledger. It is not an insurer, not an insurance intermediary and not an ancillary insurance intermediary. It has no customers of its own, holds no client money, and carries no risk. Insurance distribution is performed by the licensed distributor who integrates it. That distributor owns the customer relationship, the product approval process and the regulatory permissions. We hold none of those and do not act as if we do. Where a licensed distributor uses this platform in production, we would be an ICT third-party service provider to a financial entity, with the contractual and oversight consequences that carries under DORA. That relationship is deliberately narrow and documented rather than implied. Status What is and is not in place Item Status Insurance distribution None. RiskRouter does not advise on, propose or conclude insurance contracts, and has no customers of its own Putting cover in force Not possible through RiskRouter: nothing on this site or in the API can bind a policy FSMA registration as an intermediary Not held, and not applied for. In our own analysis, software that does not distribute needs none. What the FSMA said, and what it did not Carrier or underwriting pool None, by design: the carrier relationship stays with the distributor Professional indemnity cover Not in place Client money None is held Pricing engine A simulated rate grid while we validate; its figures are not prices anyone can buy at Evidence log Live and append-only: signed heads, public timestamps and a Bitcoin anchor, checkable without us Quote ledger Live and append-only We publish no registration number, licence reference or carrier name, because we hold none. Every price the demo shows is a simulation produced by fixed server-side rules. Frameworks Which EU rules bite, and where Listing the ones that do not apply matters as much as listing the ones that do. A platform that claims coverage of everything has usually thought carefully about nothing. Framework Relevance Where RiskRouter stands Insurance Distribution Directive Directive (EU) 2016/97 Governs anyone distributing insurance, including the conduct, information and record obligations. Applies to the distributor, not to us. The platform is built so a distributor can meet the traceability and fair-presentation obligations rather than fight the tooling. Detail below. Belgian insurance law Act of 4 April 2014, supervised by the FSMA Transposes the IDD in Belgium and requires intermediaries to be registered before distributing. No registration held , and none is sought: RiskRouter does not distribute. The registered distributor who uses the platform does, under its own registration. GDPR Regulation (EU) 2016/679 Applies to any processing of personal data. A quote record contains no personal data at all. The personal data we do hold comes from the contact form and from email sent to our published address, and from nowhere else. Set out in full on the data protection page . DORA Regulation (EU) 2022/2554 Operational resilience for financial entities, and oversight of the ICT providers they depend on. Relevant the moment a regulated distributor depends on this platform. The audit ledger, the degraded-write behaviour and the ledger health probe exist partly for this. Formal contractual arrangements would be agreed per pilot. PRIIPs Regulation (EU) 1286/2014 Key information documents for packaged retail investment products. Not applicable. These are general insurance covers with no investment component. Solvency II Directive 2009/138/EC Capital and governance requirements for insurance undertakings. Not applicable. We underwrite nothing and carry no risk. It binds the carrier standing behind the cover. ePrivacy Directive 2002/58/EC Storage of, and access to, information on a user's device. No cookie of our own, and no third-party request before consent: Google Analytics loads, and sets its cookies, only if the visitor accepts it, and refusing is as easy as accepting. The demo console keeps its engine address, and an API key only if you enter one, in your own browser. Arrivals from a tagged campaign link are counted by our own API with nothing stored on your device. Analytics , storage and campaign counts . EU AI Act Regulation (EU) 2024/1689 Obligations attaching to AI systems, including risk assessment and pricing in insurance. Not applicable. Pricing is a fixed lookup table. There is no model, no inference and no automated decision-making about any individual. Design How the IDD shaped the build Traceability, Article 20 A distributor has to be able to show what was presented to a customer and on what basis. Every routing event is appended to a ledger that records the vertical, the exact true or false state of every component, the premium and the timestamp. The database refuses edits and deletions for every role including the administrative one, so the record cannot be tidied up afterwards. Which parts of Article 20 that evidences, and which stay with you, is set out duty by duty in IDD Article 20: what you need to be able to show . Fair, clear and not misleading, Article 17 Only the server prices a policy. It discards any price a browser sends and forces the mandatory base cover back on if a modified request tries to strip it. A customer therefore cannot be shown a number the engine would not charge. The same principle governs the audit trail. When a ledger write fails, the interface says no record exists rather than implying one does. A system that reports success it cannot evidence is precisely what an inspection is designed to find. Product oversight and governance, Article 25 The rating matrix is a single frozen object with one definition per component, published at a public endpoint so the interface renders exactly what the engine charges. A change to a price is a code change with a test that fails if the console or the documentation disagrees. That gives a product owner a reviewable, versioned record of every rate that has ever applied. What the platform does not do is decide whether a product is appropriate for a target market. That judgement belongs to the distributor and the manufacturer. Division of responsibility What stays with you Using this platform does not transfer any regulatory obligation to us, and no contract we could write would achieve that. A distributor integrating RiskRouter still holds: Registration with the FSMA, or the equivalent authority in your member state The demands and needs assessment, and the insurance product information document Product approval and target market definition under product oversight and governance Professional indemnity cover and the conduct requirements attaching to your permissions The customer relationship, complaints handling and any redress Controllership of your own customers' personal data We supply routing, deterministic pricing and an audit trail you can evidence. That is the whole of it, and we would rather say so plainly than let a diagram imply otherwise. Scope Limits of this page This describes a prototype we run for technical validation. It is not legal advice, carries no contractual terms, and claims no regulated activity. Statements about EU and Belgian law are given in good faith to explain design decisions, not as an authoritative reading. Take your own advice before distributing anything. If something here does not match what the system actually does, that is a defect and we want to know. Tell us and we will correct the page or the code. ## Data protection URL: https://riskrouter.eu/privacy Data protection Written to be checked rather than believed. Everything below can be verified against the schema and the network tab, and where something is a commitment rather than a mechanism, it says so. The short version Quote records Evidence log entries Contact enquiries Email correspondence Your browser Who else touches it Your rights Who is responsible Summary The short version Pricing a quote involves no personal data at all . There is no account, no login and no field for a name. We hold personal data in two places and no others : the contact form, and the mailbox behind our published address. Both are listed below with what is kept and for how long. No cookies and no trackers unless you say yes. The site loads nothing from anyone else until you accept the analytics question; then Google Analytics counts your visit, with cookies. Refuse, or ignore the question, and nothing is sent, which you can confirm in your browser's network tab. Our own measurement is a daily count of arrivals from tagged links , which identifies nobody. Records are stored in Frankfurt, inside the EU . A daily backup of the ledger, encrypted before it leaves, is kept on GitHub for 90 days, outside the EU. It never includes the contact form. We sell nothing and share nothing for marketing. There is no advertising on this platform and never will be. Processing activity 1 Quote records A quote priced with a distributor’s API key is appended to an audit ledger, for the traceability a distributor needs under the Insurance Distribution Directive. A quote priced without a key is priced and not recorded. A record contains these fields, and none of them identify a person: Field Content id A random identifier generated by the database created_at When the quote was priced distributor_id Which distributor recorded it: a business, identified by the key it used active_vertical Which of the four risk verticals selected_components The true or false state of each cover component total_monthly_premium The premium the server computed matrix_version Which version of the rating matrix priced it compliance_status A fixed status string chain_index , prev_hash , row_hash Its place in the hash chain No IP address, no device or browser fingerprint, no session identifier and no cookie is recorded against a quote, and there is no field for the distributor’s customer. Two quotes for the same cover by the same distributor differ only in their time and their place in the chain. One use of your IP address, stated so it is not a surprise: to stop a single address flooding the engine, the API counts requests from each address over a one-minute window (60 quotes a minute) using Cloudflare’s rate limiting. That counter is kept by Cloudflare for the window and is never written to our database or joined to a quote. Because these records identify nobody, they fall outside the GDPR and are kept indefinitely as the audit trail they exist to be. The ledger refuses edits and deletions by design. Processing activity 1b Evidence log entries A distributor with a key can record the fingerprint of one of its own records, such as an advice note, in the evidence log. The record never reaches us: the distributor hashes it with a random salt it keeps, and sends only the resulting SHA-256 digest and a short tag such as idd.demands-needs . An entry holds its position in the log, when it was recorded, the distributor, the tag and the digest, and, if the distributor signs its entries, its key id, the time it claims and its signature. To us, a digest identifies nobody: we never hold the record or its salt, and a salted digest cannot be reversed. The distributor holds both, so for the distributor the digest belongs with its record and its own obligations for that record. The record stays in the distributor’s systems, where it can be erased; erase the record and its salt, and the digest left in the log cannot be linked to anyone, by anyone. Entries are kept permanently, on the same append-only terms as the quote ledger. How a digest is made Processing activity 2 Contact enquiries If you use the contact form, we store what you typed so we can reply. Nothing more is inferred, appended or enriched. Question Answer What we store Your name, email address, organisation if you give one, the enquiry type and your message What we do not store Your IP address, any tracking identifier, or anything you did not type — with one exception: if you reached the form through a link carrying a campaign tag (such as ?src=linkedin-sept ), that tag is stored with your message so we know which conversation brought you Why we may hold it Legitimate interests under Article 6(1)(f): you asked us a business question and we need to answer it How long Twelve months from your last contact, then deleted Who sees it The controller, Anil Jangra. A copy of each enquiry is emailed to our mailbox so it can be answered from there, and what activity 3 says about that mailbox applies to the copy. It is not sold or passed to anyone else Automated decisions None. Nothing about you is scored, profiled or decided by a machine The twelve-month limit is enforced by a scheduled job in the database, not by someone remembering to run a query. A retention policy nobody implements is a statement, not a safeguard. Processing activity 3 Email correspondence We publish anil@riskrouter.eu , so anything you send there is personal data we hold. Saying the form was the only such place would have been convenient and wrong. Question Answer What we store Your message and whatever your mail client puts in it: your address, your name as you send it, and the technical headers of the message Where it sits A Microsoft 365 mailbox. Microsoft is the processor for that mailbox, under its own data protection terms Why we may hold it Legitimate interests under Article 6(1)(f), the same basis as the form: you wrote to us and we need to answer How long Twelve months from our last exchange, then deleted Who sees it The controller, Anil Jangra, and Claude, an AI assistant made by Anthropic (United States), which the controller uses to read the mailbox and draft answers. A person reads every reply before it is sent, and nothing about you is decided by a machine. Anthropic provides Claude to us under its consumer terms, not under a data processing agreement ( below ). It is not sold or passed to anyone else One honest difference from the form: that twelve-month limit is a rule a person applies, not a scheduled job in a database. If the difference matters to you, use the form — it is enforced there, and we would rather you knew which of the two we can prove. Your device What we keep in your browser This site sets no cookie of its own. Google Analytics sets cookies only if you accept it, as described below . Your answer to that question is kept in your browser’s local storage (not a cookie, and never sent to us), so the question is not asked on every page. The demo console saves up to two values in your browser's local storage, so you do not have to retype them: the address of the pricing engine it should call, and, only if you enter one, your API key. The address never leaves your device. The key is sent only with requests to our own engine (or to one running on your own machine), never to any other address you type in. We cannot read either from your browser; emptying the key field removes the key, and clearing your site data removes both. The try it panel on the API reference keeps a key in the open tab’s memory only. Apart from your analytics choice and, if you accepted, Google’s cookies, nothing else is stored client-side. Google Analytics, only if you accept On your first visit a strip at the foot of the page asks whether we may count your visit with Google Analytics. Until you press Accept analytics , nothing is loaded from Google and nothing is written to your device by Google. No, thank you is as easy as accepting, and a browser sending Global Privacy Control is treated as having said no. Why To see which pages are read, which searches and sites bring visitors, and where they leave, so we can improve the site Legal basis Your consent (GDPR Article 6(1)(a), and the ePrivacy rules on storing information on your device). You can withdraw it at any time with Analytics settings at the foot of any page; withdrawing deletes the Google Analytics cookies on this site What is collected The pages you view, the page or search that brought you, the time, your browser and device type, and an approximate location derived from your IP address. Google Analytics does not log or store IP addresses of visitors in the EU. On /seal , which step you reached (a file chosen, sealed, receipt saved, a check run), never the file, its name, its fingerprint, your key or the receipt. Cookies named _ga hold a random identifier so a return visit can be recognised; they expire after two years unless you withdraw What is switched off Google signals and ad personalisation. Nothing is used for advertising, and no data from the contact form, the evidence log or an API key is ever sent to Google Who processes it Google Ireland Limited, as our processor under Google’s data processing terms. Google may process the data in the United States; Google LLC is certified under the EU US Data Privacy Framework How long Event data is kept for two months, Google Analytics’ default Campaign counts When we share a link we sometimes tag it with a campaign name, such as riskrouter.eu/integrate?src=linkedin-sept . If you arrive on such a link from another site, and stay a few seconds and scroll, point, touch or type (so that we count people, not the machines that open links in emails), the page tells our own API the campaign name and the page, and we add one to that day’s count. That is all that is sent: no IP address is stored, no identifier, no browser details, and nothing is written to your device — the tag travels in the links on the page, not in a cookie or in storage. The result is a table of numbers per day, campaign and page, which cannot be tied to any person and is not personal data. The same tag is kept with a sandbox key you take, and with a message you send, so we can tell which campaign produced a trial or a conversation. Arrive without a tag and nothing is counted at all. Sub-processors Who else touches the data Provider Role Where Cloudflare Serves the pages and runs the pricing engine Processed at the network location nearest to you, which may be outside the EEA. Nothing is stored there. Supabase Hosts the database holding quote records and contact enquiries Frankfurt, Germany GitHub Keeps the daily backup of the ledger and the distributor records, for 90 days. It never includes contact-form enquiries. The backup is encrypted (AES-256) before it is uploaded, with a passphrase GitHub does not hold, so GitHub stores it but cannot read it Outside the EU Google Google Analytics, only for visitors who accept it ( above ). It never receives the contact form, the evidence log or API keys Google Ireland Limited; possible transfer to the United States under the EU US Data Privacy Framework Microsoft Runs the mailbox behind our published address, so anything you email us passes through it Microsoft 365, under Microsoft’s own data protection terms Anthropic Claude, the AI assistant the controller works with: it reads the mailbox to draft replies, which a person reads and sends, and it can see what the controller shows it while building and running the service Anthropic, PBC, United States, under Anthropic’s consumer terms: not a data processing agreement, and not a processor acting on our instructions That is the complete list. There is no email marketing platform, no customer data platform and no advertising network, and Google Analytics is reached only after you accept it. Open your network tab and count. Microsoft appears only because we publish an email address; it touches nothing the platform itself stores. Anthropic appears because the controller reads that mailbox with Claude; we say plainly that this is under consumer terms rather than a processing agreement, because you are entitled to know which of the two it is. Request routing through Cloudflare means the processing of a request in transit can happen outside the EEA even though the stored record does not. We would rather state that plainly than claim an EU-only guarantee the architecture does not support. Your rights What you can ask for Where we hold personal data about you, which means the contact form and our mailbox and nothing else, you can ask us to give you a copy, correct it, delete it, restrict what we do with it, or object to us holding it at all. Ask through the contact form or at anil@riskrouter.eu , and we will act within one month. Because our basis is legitimate interests rather than consent, you can object at any time and we will stop unless we have a compelling reason not to. In practice, for a business enquiry, there will not be one. If you are unhappy with how we handle it, you can complain to the Belgian Data Protection Authority, the Gegevensbeschermingsautoriteit or Autorité de protection des données, or to the supervisory authority in your own member state. Accountability Who is responsible The controller of the personal data described on this page is Anil Jangra , a natural person, who builds and operates RiskRouter as a validation prototype. No company has been registered yet, so there is no legal entity, registered office or data protection officer to name here; naming one that does not exist would be worse than naming none. When a company is formed, its name and registered address will appear here before any production processing begins. Where a licensed distributor deploys this platform for their own customers, that distributor is the controller for their customers' data and we would be a processor acting on their instructions under a written agreement. Nothing on this page changes that division. For anything concerning your data, including the rights set out above, write to anil@riskrouter.eu or use the contact form . Both reach the same person, which is the whole of the organisation today. If anything here does not match what the system does, treat it as a defect. Tell us and we will fix the page or the code, whichever is wrong. ## Terms of use during validation URL: https://riskrouter.eu/terms Terms of use The terms for using this site, a free sandbox key, the API at api.riskrouter.eu and the hosted MCP endpoint at api.riskrouter.eu/mcp while RiskRouter is a validation build. Written to be read in five minutes. Last changed 24 October 2026. The short version What you may use What you send Keys What we promise, and what we do not Liability Changes Who, and which law The short version Price Free. Nothing commercial takes effect before the activity is authorised; the application is pending in Brussels. What you send A SHA-256 digest and a short tag. Never the record, never personal data. Permanence An entry can never be changed or deleted, by you or by us. That is the point of the log. Warranty None. The service is provided as it is, on a best-effort basis, without a service level. Law Belgian. What you may use Anyone may read the site, check a proof, take a free sandbox key and record digests with it, and connect an MCP client to the hosted endpoint. The verifiers, the conformance suite, the record templates and the recorders are published under the Apache License 2.0 ( Verify ); that licence, not these terms, governs what you do with their code. Do not try to get around the limits on keys and entries, to reach anything that is not published, or to make the service unavailable to others. Do not present a seal as proving more than it does (below). What you send The evidence log accepts a SHA-256 digest with a short tag (and, if you choose, a position in a completeness chain or your own signature over them), and refuses a record or anything that is not a digest. Hash your own salted record and keep the record and the salt yourself. Do not put personal data in a tag, a chain tag or anything else you send: whatever you send is kept for as long as the log exists. What we hold, and why, is on Data protection . Keys Keep your key secret: anyone holding it can record entries in your name. We store only its SHA-256, so we cannot show it to you again. We may revoke a key that is used against these terms; the entries it recorded stay, because no entry is ever removed. What we promise, and what we do not We promise that no entry is ever changed or removed, that every signed head and proof can be checked with the published keys and open-source verifiers without asking us, and that an answer is never shown as a success it is not. We do not promise availability, a response time or that the service will keep running unchanged. A seal shows that a record is the one that existed when it was sealed; it does not show that the record was true or complete, or that anyone met an obligation. Nothing on this site or in an answer of the API is legal advice. Liability The service is free and provided as it is during validation. To the extent Belgian law allows, we are not liable for indirect loss, lost profit or lost data arising from its use or from its being unavailable. Nothing in these terms limits a liability that the law does not allow to be limited. Changes Any change to these terms is dated on this page and listed on the changelog . Paid use, once the activity is authorised, will be under a written contract, never under a silent change to this page. Who, and which law RiskRouter is built and operated by Anil Jangra , in Brussels, as a validation prototype. No company has been registered yet; when one is, its name and registered address replace this paragraph. Questions: anil@riskrouter.eu . These terms are governed by Belgian law, and the courts of Brussels have jurisdiction, unless a rule that protects you as a consumer says otherwise. ## Contact RiskRouter URL: https://riskrouter.eu/contact Contact Write to us about a pilot, an integration question, or something on the regulatory pages you think is wrong. A person reads every message. There is no sales sequence and no newsletter. Send a message Your name Email Used to reply to you. Nothing else, ever. Organisation (optional) What is this about Running a pilot Integration or technical question Regulatory or data protection Something else Message At least ten characters. Leave this field empty Send message We store your name, email, organisation and message so we can reply, on the basis of legitimate interests, and delete them twelve months later. That deletion runs on a schedule in the database rather than depending on anyone remembering. How we handle it . What to expect Straight answers If you are evaluating this for a checkout, say what you sell and roughly what volume. That is usually enough for us to tell you whether it fits, including when it does not. If you found something wrong on the compliance or data protection pages, quote the line. Those pages are meant to be checkable, and a correction is worth more to us than a compliment. Before you write Two things worth knowing We are not selling insurance. RiskRouter is infrastructure for licensed distributors. If you were looking to buy cover for yourself, we cannot help and would rather say so now than take your time. RiskRouter holds no FSMA registration and does not distribute insurance, so a pilot means building and testing with your product, under your own registration. The regulatory position spells out where the line sits, and the pilot letter what each side commits to. The other channel Or write directly anil@riskrouter.eu The form is still the better route, and not for our convenience: it writes straight to a database in Frankfurt under a twelve-month deletion schedule you can read in full. An email sits in a mailbox instead, so it is governed by a rule someone has to apply rather than one the database enforces. Both are covered on the data protection page ; only one of them is automatic. No legal entity is registered yet, so none is named here. When one exists its details will appear on the data protection page . ## Verify the ledger URL: https://riskrouter.eu/verify Verify the ledger yourself Every platform in this market says its audit trail can be trusted. Ours can be checked. Record the hash below today, and you can prove later that not one historic quote was altered, without asking us for anything and without taking our word for it. Live Current head of the chain Read straight from the running ledger. It changes with every quote priced. … Entries chained SHA-256 Digest … As of (UTC) Head hash loading… Copy head hash Keep it with today's date. That single string commits to every entry written so far. How it works Each entry seals the one before it When a quote is priced, the database computes a SHA-256 digest over that entry's contents and the previous entry's digest , before the row is written. Change any field of any historic entry and its digest no longer matches, and neither does every digest after it. The break is visible at the exact entry. The canonical form is deliberately plain, so any language can rebuild it: sha256( chain_index | id | created_at | vertical | components(key=true|false, sorted by key, comma separated) | premium(0.00) | status | matrix_version | distributor_id | prev_hash ) No JSON formatting, no key-order dependence, no locale. The first version of this used PostgreSQL's own JSON output and was replaced, because a chain only one database can rebuild is a claim rather than evidence. distributor_id is inside the digest, not beside it. Attribution that sat outside the hash could be reassigned after the fact with every digest still verifying — which would make the chain evidence of what was quoted but not of who quoted it. Do it yourself Four commands, none of which need us The verifier shares no code with the database. It rebuilds every digest in Node. # 1. today: save the whole signed attestation, not just the hash curl -s https://api.riskrouter.eu/api/v1/ledger/attestation > 2026-09-20.json node tools/verify-attestation.mjs 2026-09-20.json # 2. any time: export your own entries and the chain they sit inside curl -s -H "Authorization: Bearer $RISKROUTER_API_KEY" \ https://api.riskrouter.eu/api/v1/ledger/export > export-2026-09-20.json # 3. check it. the first proves we signed this head, the second proves # your entries are what produces it node tools/verify-attestation.mjs export-2026-09-20.json node tools/verify-ledger.mjs export-2026-09-20.json # 4. months later: a head you recorded must still come out of history node tools/verify-ledger.mjs export-2027-03-01.json --expect-head # 5. to show one quote to a regulator or a customer, and no other node tools/entry-proof.mjs export-2026-09-20.json 42 > entry-42.json node tools/verify-ledger.mjs entry-42.json A mismatch at step 4 means history changed. There is no version of that sentence where we get to explain it away. Step 5 cuts a single-entry proof : that entry in full, its links to the head, and the signed attestation. It proves the entry’s content produces the digest the ledger published for its position. The links after it are followed by digest only, as in any export; they bind because anyone who later recomputes the full ledger must reach the signed head, and cannot if that entry was changed or removed. You can also make one on this page, below, after checking an export. Step 2 used to read differently. It told you to pull the rows out of the database with a Supabase key — a credential you will never hold, because ours never leaves the Worker. So the independent check this page promised was one no customer could actually run. A proof that needs the operator’s cooperation is not a proof, it is a courtesy, and we would rather say that plainly than leave the old instruction sitting there looking reassuring. Save the whole response rather than copying the hash out of it. The signature is what turns a string you typed into a commitment we made: it covers the head hash, the entry count and the time together, so none of the three can be changed afterwards without the signature failing. The public key is in anchors/signing-key.json in the repository. Checking against a key we hand you at the moment you check would prove nothing, if we are the thing in question. Check a file here Drop in your export. It is checked in this browser, not by us. The same checks as the commands above, run on this page: the signature on the head, every link in the chain, and each of your own entries rebuilt from its content. The file is read by your browser and never sent anywhere — this section makes no network request at all, and a test in the repository fails if one is ever added. Choose a file or drop it here An export from /api/v1/ledger/export , a single-entry proof, a saved attestation, or the rows downloaded from the demo below. Head you recorded earlier (optional) Show one quote without showing the rest. Pick one of your entries and download a single-entry proof: that entry in full, its links to the head, and the signed attestation. Whoever you give it to can check it here or with the public verifier, and learns nothing about your other quotes. Download proof Which public key is this checked against? The one published in anchors/signing-key.json and in the public verifier repository (whose tools are Apache-2.0), key id . Be clear about what that means: this page is served by us, so the key it carries is also one we handed you just now. If you saved a copy of the key earlier, load it here and the check uses yours instead. Use my own copy of the key using the key embedded in this page The same reasoning applies to the code: a page we serve could, in principle, be made to say INTACT regardless. That is why this is a convenience and the commands above are the proof — they run code you hold, against a key you hold, and share nothing with this site. Break it yourself Change a premium and watch the chain refuse it Four entries, chained exactly as the real ledger chains them — same canonical form, same SHA-256, computed in your browser. Edit any premium below. The entry you touch stops matching its own digest, every entry after it stops linking, and the head moves. Nothing here talks to us, and nothing you type leaves the page. # Vertical Premium /mo Digest Status Head after your edits: Download these rows Reset the chain The download is the part worth doing. Drop it into the checker above , or run the independent verifier over it, and both reach the same conclusion this page did — including the entry number where it broke: node tools/verify-ledger.mjs rows.json Two implementations that share no code agreeing on where history was altered is the whole argument. A test in this repository asserts that the function running in your browser and the one in tools/verify-ledger.mjs produce byte-identical canonical strings, so the demonstration cannot quietly drift away from the thing it demonstrates. Continuity What you keep if RiskRouter stops existing RiskRouter is run by one person. You should ask what happens to your audit trail the day it stops, and the answer should not depend on our goodwill on that day. An export you have already downloaded keeps working after we are gone. It contains: Every one of your entries in full, so each digest can be rebuilt from its own content. The (chain_index, row_hash, prev_hash) of every entry in the ledger — opaque digests, no one else’s content — which is what links your entries to the head. The signed attestation for that head, and the id of the key that signed it. The verifier is forty lines of standard-library Node in a public repository, the public key stays in anchors/ including after any rotation, and neither needs a server to run. Download an export once a month and you hold evidence that outlives the company that issued it. That is the opposite of lock-in, and it is deliberate: a product whose value depends on you being unable to leave is a product we would not want to sell. The full continuity and exit plan covers what stops, how to leave, and the facts your DORA register needs. What a copy cannot reproduce How old the record is 13 September 2026 first head anchored 16 heads anchored 967 830 earliest Bitcoin block 46 entries that head covers none yet independent witnesses 38 heads with RFC 3161 time Everything else on this site could be rebuilt by someone else. These figures could not. They are computed, each time the site is built, from the signed anchors and timestamp proofs in anchors/ and the public keys in witnesses/ , and from nothing else; the same figures are published as proof-of-age.json . The 13 September 2026 head is in Bitcoin block 967 830: whatever anyone says later, including us, it existed before that block was mined. Every day the evidence log moves, its newest signed head goes into Bitcoin as well: 15 heads are anchored so far, 15 already in a block, the earliest block 972 079. 38 heads carry an RFC 3161 time-stamp token from DigiCert timestamp service and Sectigo timestamp service, the latest issued at 2026-11-14 15:09:04 UTC. None is a qualified eIDAS time stamp: they are evidence of time from an independent authority, without the legal presumption a qualified one carries. No independent party co-signs the evidence log yet, and the page says so rather than leaving the figure out. Every head, day by day How to become a witness . Honest limits What this does and does not prove Claim Status A historic entry was altered or deleted Detectable by anyone holding an earlier head Entries were re-ordered or a gap introduced Detectable An entry was fabricated and inserted into the past Detectable A quote was never written at all Not detectable. A chain proves what is there, not what is missing The premium was correct when it was quoted Separate control: the rating matrix is versioned and stamped on every entry Someone rebuilt the whole chain from scratch Detectable against any head recorded beforehand. node tools/verify-anchors.mjs export.json checks a ledger against every head published in anchors/ , and a rebuild cannot reproduce one taken before it. Residual limit: that directory is ours, so a rebuild plus a force-push could remove the anchor too — not from a copy you already hold, which is the reason to keep one We deny ever having published the head hash you hold Every attestation is signed. One you saved is ours whether we like it or not A head hash was published before the entries it covers existed Bounded, for anchored heads. The 13 September 2026 anchor was submitted to three independent OpenTimestamps calendars on 20 September, and those proofs are now upgraded to Bitcoin attestations for blocks 967830, 967831 and 967860. So that head, and the 46 entries it covers, existed no later than block 967830 was mined. It proves an upper bound, not the exact moment: the time between an entry and the next anchored head is still our word. Check a proof yourself with ots verify , which compares it with the block’s header; the files are in anchors/ The last row is the honest one, and it is the reason the row above it matters. Publishing the head is the commitment; a signature means we cannot later say we never made it; keeping your own copy is what makes it binding. What none of that establishes on its own is when . Bitcoin does that for anchored heads: a head timestamped there existed before that block was mined, whatever we say later. Heads are anchored every day the log moves, so the window in which a time is only our word is at most about a day plus the time a block takes to confirm; the RFC 3161 tokens, taken every two hours, narrow it to about two hours, on the authority’s word. Every head, day by day: /age . ## IDD Article 20: what you need to be able to show URL: https://riskrouter.eu/guide-idd-article-20 Guide for insurance distributors IDD Article 20: what you need to be able to show Article 20 of the Insurance Distribution Directive, Directive (EU) 2016/97, is about the sale itself: what the customer needs, and whether what you proposed fits. It never uses the word record . But every duty in it is one you may be asked to demonstrate long after the conversation is over, and the only way to demonstrate it is with evidence you kept at the time. The text What Article 20 asks of a distributor Before an insurance contract is concluded, the distributor must: Specify the customer’s demands and needs , on the basis of information obtained from the customer. Give objective information about the product in a comprehensible form, so the customer can make an informed decision. Propose only what fits. In the directive’s words: Any contract proposed shall be consistent with the customer’s insurance demands and needs. Where advice is given, explain it : a personalised recommendation explaining why a particular product would best meet the customer’s demands and needs. For non-life products, hand over the standardised insurance product information document (the IPID). Article 23 then governs how that information reaches the customer: on paper, or on another durable medium or a website where the conditions for that are met. In Belgium the directive is transposed by the Act of 4 April 2014 on insurance, and the FSMA supervises its conduct rules. Evidence Each duty, and what would show you met it A duty you cannot evidence afterwards is, in an inspection or a complaint, a duty you cannot show you met. This is the practical reading: for each part of Article 20, the question someone will eventually ask, and what answers it. The last column says honestly which parts a system like RiskRouter can evidence and which stay entirely with you. Duty The question you will be asked Who holds the evidence Demands and needs What did the customer tell you, and what did you conclude they needed? You. RiskRouter records no customer information at all, by design. Consistency of what was proposed Exactly what was proposed: which product, which options, at what price, when? RiskRouter records this: the vertical, the on or off state of every component, the premium, the rating-matrix version and the time, in an entry nobody can edit afterwards. Objective information Was the price you showed the price the rules produced, and which rules were they? Partly. The server computes every price and ignores any a browser sends, and each entry carries the version of the rate table that priced it. The wording you showed stays in your own system. Personalised recommendation Where you advised, what did you recommend and why? You. IPID (non-life) Which version of the product information document did the customer receive? You, and your product manufacturer. Durable medium (Article 23) How and when did the information reach the customer? You. How long to keep this evidence is not something this page states. Check the period that applies to you with your supervisor or your professional federation. The weak point Evidence that could have been edited proves less Most distributors do keep records: a CRM note, a spreadsheet, a PDF in a shared drive. The difficulty is that every one of those can be changed after the fact by anyone with access, and usually leaves no trace when it is. A record that could have been corrected is a record whose accuracy rests on your word, which is exactly what it was meant to replace. That is the gap an append-only, hash-chained ledger closes: changing any entry breaks every entry after it, and anyone holding an earlier checkpoint can see that it happened. The difference is set out in database log vs tamper-evident ledger , and you can watch a chain refuse an edit on the verify page . Scope What this page is not This is a plain-language reading of the directive, written to explain why RiskRouter records what it records. It is not legal advice and not an authoritative interpretation; your own counsel, your supervisor and your federation are. Using RiskRouter transfers none of your obligations to us. The regulatory position sets out what stays with you in full. Get a sandbox key The audit-trail checklist ## Audit-trail checklist for insurance brokers URL: https://riskrouter.eu/guide-audit-trail-checklist Guide for brokers and distributors Audit-trail checklist for insurance brokers Twelve questions to put to any system that holds the record of what you quoted: your CRM, your broker software, a spreadsheet, or us. They are written for a broker in Belgium, where the FSMA supervises the conduct rules of the Insurance Distribution Directive, but nothing in them is Belgian-specific. Our own answer follows each one, including where it is no. 1 3 What is in the record Does it hold what you would be asked for? Question Why it matters RiskRouter 1. Does each entry say exactly what was proposed? Travel cover, about 6 does not show the proposal fitted the customer’s needs. Every option, on or off, does. Yes. The vertical, every component’s on or off state, and the premium in integer cents. 2. Does it record which price rules applied? A price changes. Without the version, a two-year-old quote cannot be explained. Yes. Every entry carries the rating-matrix version that priced it. 3. Does it record the demands and needs assessment? It is the heart of IDD Article 20. See what Article 20 asks . No. We hold no customer information, so that assessment stays in your own system. 4 7 Integrity Could the record have been changed? Question Why it matters RiskRouter 4. Can anyone edit or delete an entry, including an administrator? If an administrator can, the record is only as good as that administrator’s word. No one can. The database refuses edits and deletions for every role, including the administrative one. 5. If an entry were changed, would anyone be able to tell? Preventing edits is a policy. Detecting them is evidence. Yes. Entries are hash-chained: changing one breaks every entry after it. Try it. 6. Can the vendor deny a checkpoint it gave you? A checkpoint the vendor can disown protects the vendor, not you. No. Every attestation of the chain’s head is signed; one you saved is ours whether we like it or not. 7. Is the time of each entry independently proven? Without it, when is still the vendor’s word. Bounded. Our 13 September 2026 head is timestamped in Bitcoin (blocks 967830 and later, via three OpenTimestamps calendars), so every entry it covers existed before that block. The exact time between an entry and the next anchored head is still our word; heads are anchored every day the log moves ( every head, day by day ). 8 10 Independence Can you check it without the vendor? Question Why it matters RiskRouter 8. Can you export your own entries at any time? Evidence you can only see inside the vendor’s screen is evidence the vendor controls. Yes, one authenticated request, no one to ask. 9. Can you verify the export without the vendor’s software? If the checker is theirs, the check is theirs. Yes. Dependency-free verifiers in a separate public repository , or the checker on the verify page , which runs in your browser and sends nothing. 10. Does the evidence survive the vendor going out of business? The record has to outlive the contract. Yes. An export you already hold verifies with no connection to us. 11 12 Honesty and data protection Does it tell the truth when it fails, and keep personal data out? Question Why it matters RiskRouter 11. When a write fails, does the system say so? A system that shows saved for a record that does not exist is the failure an inspection is designed to find. Yes. A quote whose entry was not written is reported as not recorded, never as a success. 12. Is personal data kept out of the permanent record? A record nobody can edit is also a record nobody can erase. Under the GDPR, personal data in it would be a problem you cannot fix later. Yes. A quote entry has no field for a name, an email or an identifier. See data protection . One thing no system of this kind can prove: that every quote you gave was sent to it. A quote that was never recorded leaves no trace to find. A chain proves what is in it, not what is missing, and anyone who tells you otherwise is describing something else. Scope How to use this list Use it on any supplier, including us, and ask for the evidence behind each answer rather than the answer. It is a practical checklist, not legal advice, and it does not say how long you must keep records: check that with your supervisor or your professional federation. Check our answers with a sandbox key Log vs ledger, explained ## Database log vs tamper-evident ledger URL: https://riskrouter.eu/guide-database-log-vs-ledger Guide for compliance, audit and engineering Database log vs tamper-evident ledger Almost every system that sells insurance keeps an audit log. Almost none of those logs could show that they have not been edited. That is not a flaw in any particular product; it is what a database table is. This page explains the difference in plain terms, using our own ledger as the worked example, and says what a ledger still cannot prove. The ordinary case What an audit log in a database is A table with one row per event: who, what, when. It is genuinely useful, and it answers most day-to-day questions. Its limit is that every row is just data. Whoever can write to the database, whether an administrator, a support engineer, a migration script or someone restoring a backup, can change a row, delete one or insert one in the past, and the table afterwards looks exactly as if it had always been that way. Logging the changes to the log does not help, because that second log is also a table. The record is as trustworthy as the least careful person with access to it, which is why a regulator or a court may treat it as your account of events rather than as evidence of them. What changes What makes a ledger tamper-evident 1. Each entry is sealed to the one before it Every entry is reduced to a fixed text form and hashed with SHA-256, and that hash includes the previous entry’s hash. Change one premium in one old entry and its hash changes, so the next entry no longer links to it, and nor does anything after. The newest hash, the head , therefore stands for the entire history. 2. The database refuses to rewrite it In our ledger, updates and deletions are refused by the database itself, for every role including the administrative one. That is prevention. The chain is what makes it more than a policy: if the refusal were ever bypassed, the chain would show it. 3. The head leaves the building A chain only proves something against a head someone else already holds. Ours is published in a signed attestation that anyone can fetch and keep, and past heads are committed to the public verifier repository . Once you hold one, any later rewrite of the history before it is detectable by you, without asking us. 4. The head is signed A signature means a head you saved cannot later be disowned as something we never published. You can check one in your browser on the verify page , or with the scripts in the public verifier . Side by side The same questions, asked of each Question Audit table Tamper-evident ledger Can an administrator change an old entry? Yes, silently Refused by the database, and detectable if forced Could an entry be inserted into the past? Yes, silently Detectable Could entries be reordered or one removed? Yes, silently Detectable Could the whole history be rebuilt from scratch? Yes Detectable against any head recorded beforehand Can someone outside check it? Only by trusting a screen or a copy Yes, from an export, offline Does it prove a quote that was never written? No No. A chain proves what is there, not what is missing Does it prove when each entry was made? No An upper bound, once a head is timestamped. Our 13 September 2026 head is in Bitcoin block 967830 and later, so its entries existed before that block. The exact moment is still our word Trade-offs What it costs you Nothing in a ledger can be corrected, only followed by a new entry. That is the point, and it has a consequence: personal data must never go into it, because under the GDPR a person can ask for their data to be erased and an entry that cannot be changed cannot be erased either. Our quote entries have no field for a name, an email or an identifier for exactly that reason; the personal data you hold about a customer stays in your own system, where it can be deleted. Scope What this page is not A technical explanation, not legal advice. Whether a given record meets a given obligation is for you, your counsel and your supervisor to judge. What we can offer is a record whose integrity you can check yourself, which is a better starting point for that judgement than one you have to take on trust. Watch a chain refuse an edit Get a sandbox key ## What the AI Act asks of your logs URL: https://riskrouter.eu/guide-ai-act-logs Guide for engineers building AI features in the EU What the AI Act asks of your logs If you build or run a system that the EU AI Act classes as high-risk, a handful of its articles decide what your logs must do. This page reads them as an engineer would, from the consolidated text that includes the amendment of July 2026, quoting the words that matter. It also says what the Act leaves to you, and how to keep logs that someone will believe years later. It is a reading, not legal advice, and whether your system is high-risk is yours to establish. First, the dates From when The logging, oversight and retention duties below sit in Chapter III, Sections 2 and 3 (Articles 12, 14, 19 and 26). Regulation (EU) 2026/1744 amended Article 113, which now applies those sections: 2 December 2027 to AI systems that are high-risk under Article 6(2) and Annex III (the list of uses: credit scoring, recruitment, access to essential services and the rest); 2 August 2028 to AI systems that are high-risk under Article 6(1) and Annex I (AI that is, or is a safety component of, a product covered by the EU product legislation listed there). The Regulation as a whole applies from 2 August 2026, with other exceptions listed in Article 113. A system designed now will still be running in December 2027, so the logging it needs is a design decision today, not a retrofit later. What the Act asks Six obligations, in the Act’s words Article 12(1) High-risk AI systems shall technically allow for the automatic recording of events (logs) over the lifetime of the system. Logging is a property of the system, built in, not a procedure someone remembers to follow. Article 12(2) The logs must enable recording of events relevant for identifying situations that may make the system present a risk or amount to a substantial modification, for the provider’s post-market monitoring (Article 72), and for the deployer’s monitoring of its operation (Article 26(5)). That is a test of purpose, not a field list: each event you record should serve one of the three. Article 12(3) Only for remote biometric identification (Annex III, point 1(a)) does the Act name fields: the period of each use, the reference database, the input data that led to a match, and who verified the result. Articles 19 and 26(6) Providers and deployers keep the logs under their control for a period appropriate to the intended purpose of the high-risk AI system, of at least six months , unless other EU or national law, in particular on personal data, provides otherwise. Financial institutions keep them as part of the documentation their financial-services law already requires. Article 14(4)(d) The people overseeing the system must be able to decide, in any particular situation, not to use the high-risk AI system or to otherwise disregard, override or reverse the output . If an override can happen, it is an event worth recording, with who did it and why. Article 86(1) A person affected by a decision taken on the basis of an Annex III system’s output can obtain clear and meaningful explanations of the role of the AI system in the decision-making procedure and the main elements of the decision taken . You can only explain one decision months later if you recorded that decision, not just the request that led to it. What it leaves to you What the Act does not say It prescribes no format, no schema and no storage. Outside remote biometric identification it names no fields. It does not say logs must be tamper-evident, held by a third party or signed. Six months is a floor, and the GDPR pulls the other way: personal data in a log is personal data, kept no longer than needed. So the Act asks you to have logs. Whether anyone believes them later, when a supervisor, an auditor or a court asks what the system did in one case, is a different question, and the Act leaves it to you. A log that its own administrators could have edited is your account of events, not evidence of them. For the engineers Logs that hold up 1. Record the decision, not just the call For each decision: the system and its version, a reference to the case (never a name), a digest of the input and the output exactly as they were, the output in words, the role the output played, the decision taken, and whether a person reviewed it. For each oversight action: who (by reference), what they did (confirmed, overrode, reversed, did not use, stopped) and why. That is what Articles 14 and 86 will ask you to show. 2. Keep personal data out of the log where you can Put references and SHA-256 digests in the log and keep the content in your own store with its own retention. The digest still proves the content later, and the log does not become a second copy of personal data to manage. 3. Make it append-only where you keep it Write-once storage, or a database that refuses UPDATE and DELETE on the log table by trigger, so a mistake or a bad day cannot rewrite history quietly. 4. Make it evident to people who are not you Chain the entries by hash or put them in a Merkle tree, then publish the head somewhere you do not control: a time-stamping authority (RFC 3161), independent witnesses, or a public log. Then even your own administrators cannot change an old entry without it showing. This is the part that turns a log into evidence. 5. Check that you can prove an old entry Pick an entry from six months ago and prove, from what you kept, that it is the one written then. If you cannot do that today, you will not be able to when someone asks. How to do it Yourself, or with an existing log You can do all of the above without us. Open-source transparency-log software such as Trillian or Tessera gives you the Merkle tree; free RFC 3161 authorities will time-stamp a head; write-once object storage covers the rest. It is work, but it is ordinary engineering. Or keep the records where they are and seal only their salted digests in an existing log. RiskRouter’s recorders are Apache-2.0 and have no dependencies; the record and its salt stay in your journal, and only the digest and a kind are sent. With the AI Act pack ’s kinds: pip install riskrouter-recorder import os from riskrouter_recorder import Recorder recorder = Recorder(api_key=os.environ.get("RISKROUTER_API_KEY"), journal="evidence.jsonl") recorder.record({"pack": "ai-act", "pack_version": 1, "kind": "ai.decision", "record_id": "dec-2026-118305", "system_ref": "credit-scorer", "system_version": "4.2.1", "case_ref": "app-552091", "output_summary": "Score 612 of 1000; band C.", "role_of_system": "The score set the band; an analyst confirmed it.", "decision": "Credit offered at EUR 5 000 at the band C rate.", "human_involved": True, "decided_at": "2026-09-26T11:42:10Z"}) Before installing anything, you can see the whole round trip on one file of your own at /seal : the file stays in your browser, and its receipt is checked by the same offline verifiers. Without a key, the recorder keeps the journal and sends nothing, so you can try it locally first. The integration page has the Node version, the logging handler, the decorator and the OpenTelemetry, LangChain and OpenAI Agents adapters; the specification and the offline verifiers show how a receipt is checked without us. What a sealed log still does not prove That the log is complete A seal proves an entry was not changed, not that nothing was left out. Numbered chains make a gap visible ( completeness ), and only for what was chained. That the decision was right It proves what was decided and when, not whether it was fair or lawful. That you meet the Act Logs are one obligation among many (risk management, data governance, documentation, accuracy, human oversight). Recording does not make a system compliant with anything. Quotations are from the consolidated text of Regulation (EU) 2024/1689 that includes Regulation (EU) 2026/1744, on EUR-Lex ; only the Official Journal is authentic. A weekly check compares every article cited here with EUR-Lex and flags any change. ## Security and vulnerability disclosure URL: https://riskrouter.eu/security For security reviewers and researchers Security and vulnerability disclosure If you have found a way to break something here, we want to hear it, and we would rather hear it from you than read about it. This page says where to write, what is in scope, what we promise in return, and which controls are already in place, so you can check them rather than take them on trust. Report How to report Email anil@riskrouter.eu with security in the subject. Say what you found, how to reproduce it, and what you think the impact is. English or French. The same contact is published at /.well-known/security.txt . RiskRouter is run by one person, so you will be writing to the person who wrote the code. We aim to acknowledge a report within five working days and to tell you what we are doing about it. We will credit you when it is fixed, if you want to be credited. There is no bug bounty: we have no money to offer, and would rather say so than imply otherwise. Scope What is in scope riskrouter.eu : the site and the demo console. api.riskrouter.eu : pricing, recording, the ledger export and attestation, the evidence log and its proofs, checkpoints, the MCP endpoint, sandbox keys. The ledger’s guarantees : any way to change, insert, reorder or delete an entry without it being detectable, to get an attestation signed that we did not produce, or to make a failed write look like a success. Attribution : any way to record against a distributor without that distributor’s key, or to read another distributor’s entries. The public verifier at github.com/aniljangrabe-dev/riskrouter-verify : any input that makes it report a tampered ledger as intact. Please do not Test for denial of service or send floods of requests. The ledger’s ceilings are deliberate and documented. Try social engineering, or anything involving our email provider or hosting providers’ own systems. Record personal data in the ledger. Entries can never be deleted, so anything you put there stays forever. A sandbox key is enough to test recording. Access, keep or share data that is not yours beyond what you need to show the problem. Research that follows this page, in good faith, is welcome. We will not pursue legal action against it. Controls What is already in place, and how to check it Control What it does Check it Content-Security-Policy Every page may run only the scripts it shipped with, listed by hash. No inline code that was not there at build time, no framing, and one third-party script origin: Google’s analytics tag, loaded only after the visitor accepts. Response headers of any page No third-party requests before consent The site loads nothing from any other origin, no CDN and no font service. Google Analytics loads only if the visitor accepts it. Your browser’s network tab Transport and framing HSTS, nosniff , framing denied, a strict referrer policy, on the site and the API; the site also isolates its windows from other origins ( Cross-Origin-Opener-Policy ) and turns off camera, microphone, location, payment and USB access. Response headers Keys stored as digests Only the SHA-256 of an API key is ever stored. A copy of our database would contain no usable key. API reference Append-only ledger The database refuses edits and deletions for every role; entries are hash-chained and the head is signed. Verify page One outbound destination The API Worker contacts nothing but its own database. An automated test fails the build if a new destination appears. Trust page A pinned, scoped pipeline Every GitHub Action in the build, deploy and scheduled jobs is pinned to a full commit SHA, which cannot be moved the way a version tag can. Every workflow declares the least permission its token needs, and the one that deploys can only read the repository. No workflow runs a fork’s code with secrets. A test fails the build if any of this slips. Questionnaire Deploys checked from outside After every deploy the pipeline asks the live API which commit it runs, and fetches every page in the sitemap to compare its title with the build’s. A deploy that did not land fails instead of passing quietly. Status , GET /api/v1/health Admin behind an identity check The admin console sits behind Cloudflare Access and verifies each session’s signature. It can do five things, none of which touch a quote or the ledger. Not publicly reachable, by design Honestly What we have not done No external penetration test and no security certification. We claim neither. The controls above are what a one-person validation build can put in place and let you check; if a regulated distributor needs an independent assessment before a pilot, that is a reasonable thing to ask for and to plan together. Sending us your vendor questionnaire? Most of it is already answered, with evidence, on the security questionnaire page. ## Continuity and exit plan URL: https://riskrouter.eu/continuity For procurement, risk and DORA review Continuity and exit plan RiskRouter is run by one person, Anil Jangra, and has no legal entity registered yet. If you depend on it, you should know what happens the day it stops, and the answer should not rest on our goodwill that day. DORA asks a financial entity to have an exit strategy for each ICT provider it relies on; this page is the part of that strategy we can write for you, and it says where it falls short. What you keep Your evidence outlives us Your entries, whenever you want them. One authenticated request to GET /api/v1/ledger/export returns every entry you recorded, the digest skeleton of the whole chain they sit in, and the signed head. No ticket, no notice, no fee. A verifier we do not control the outcome of. Dependency-free verifiers in a public repository that shares no code with the service. Clone it now and it keeps working without us. The public key and past heads. Published in that same repository, so an export you hold can be checked against a key and a head you did not get from us on the day. An open format. The export is JSON, and the canonical form of an entry is documented on the verify page , so any engineer can re-implement the check. What makes this real is the copy you take. An export saved regularly, kept somewhere we cannot reach, is evidence that survives RiskRouter. One you meant to download is not. What stops What stops if we stop Function If RiskRouter stops What you need in place Pricing The API stops answering. A fallback for quoting, or your own copy of the engine (see below). Recording new quotes Stops. A quote after that point is not in our ledger. Your own record of new quotes from that date. Your existing evidence Unaffected, if you hold an export. Regular exports, stored by you. Checking that evidence Unaffected. A clone of the public verifier. Backups Restore tested, and what backups exist A restore drill runs every day, and every run so far has passed. Each day a backup of the production ledger and the evidence log is taken and restored into a fresh database, automatically; a failed drill alerts the owner and is written up on the changelog . The first drill was run by hand on 23 September 2026, and the job has run on a schedule since. In the first drills, all 46 entries came back, the database’s own chain check and an independent rebuild both reached the same head, that head is the one signed on 13 September and timestamped in Bitcoin, and the restored ledger still refused every edit and deletion. Loading took under a second; a real recovery would take as long as noticing, deciding and repointing the service. What is not in place yet: the database runs on a plan that keeps no backups of its own. The only backups are our daily ones, kept encrypted on GitHub for 90 days, so entries recorded since the last one, at most a day’s, would be lost with the database. Your own exports are not affected: they verify without us. Exit The steps to leave, in order Take a final export and verify it with the public verifier, offline. Save the signed attestation it contains, and a copy of the public key from the verifier repository. Stop sending your key; record new quotes in your own system from that date. Ask us to revoke your key, so nothing more can be recorded in your name. Keep the export for as long as your own retention obligations require. None of these steps needs our cooperation except the revocation, and even without it an unused key records nothing. That is deliberate: an exit that depends on the provider is not an exit. Running it yourself It runs in your own account, and the source is not public today The engine is small: a SQL schema and one Worker, with no proprietary runtime. There is a deployment kit for running that same code in your own account: on your own Cloudflare and Supabase, or anywhere containers run (PostgreSQL, PostgREST and the API, started with one command). Your instance signs its heads with its own key, holds its own data, and calls nothing of ours: no licence server, no telemetry. Anchoring, witnesses and the conformance suite work against your instance the same way they work against ours. The kit is built and exercised end to end from nothing every week in our CI: a distributor and a key, a quote recorded, evidence recorded, and every signature checked with the instance’s own key by both verifiers. The engine’s source code is still in a private repository; what is published, under the Apache License 2.0, is everything that checks or feeds it: the specification and test vectors, both verifiers, the conformance suite, the integration kit and the drop-in recorders. So running it yourself needs the source and a licence from us, agreed in writing. If that is a condition of your pilot, say so before it starts, and agree how you would get the source if we stopped: direct access, or an escrow arrangement. We would rather settle that at the start than have you discover it at the end. DORA Facts for your register of information A financial entity keeps a register of its contractual arrangements with ICT third-party service providers. These are the facts about us you will need for it. The ones that count against us are stated as plainly as the rest; whether the function we support is critical or important is your assessment, not ours. Item RiskRouter Provider RiskRouter. No legal entity is registered yet, so there is no LEI and no company number. A contract can only be signed once one exists. Service Software delivered as an API: deterministic server-side pricing from a published rating matrix, and an append-only, hash-chained audit ledger with a signed head. Function it supports Pricing configurations, and recording what was quoted as evidence for your IDD obligations. It does not hold the customer relationship, the demands-and-needs assessment, the IPID, claims or payments. Personal data None in a quote entry, by design. Enquiries sent through our contact form are the only personal data we hold, and they are not part of the service. Where data is stored Supabase, Frankfurt, Germany (EU). An encrypted daily backup is kept on GitHub for 90 days, outside the EU. Where it is processed Requests are handled by Cloudflare at the network location nearest the caller, which may be outside the EEA. Nothing is stored there. Subcontractors Cloudflare (network and compute), Supabase (database), GitHub (encrypted backups), Microsoft (the mailbox behind our published address). Listed on the data protection page . Substitutability Your data leaves in full, in an open format, verifiable without us. Pricing and recording would need replacing: see what stops, above. Exit plan This page. Certifications and audits None. No penetration test and no certification is claimed. See security . Service level, incident notice None offered today. To be agreed in writing per pilot. Contract terms Each provision DORA Article 30 requires, and the clause we offer for it: DORA Article 30, clause by clause . Status Validation build. Mock pricing, no carrier, no cover in force. See the regulatory position . Try the export with a sandbox key Discuss a pilot’s terms ## Changelog and rating matrix history URL: https://riskrouter.eu/changelog For product owners and integrators Changelog and rating matrix history What changed, and when. The rating matrix comes first, because a quote recorded today may need explaining in two years, and the explanation starts with which prices applied. Everything below the matrix is taken from the repository’s own history. Rating matrix Every version of the prices Every recorded entry carries the matrix_version that priced it, and that field is inside the entry’s hash, so it cannot be changed afterwards. The fingerprint is the SHA-256 of every price, one line per component ( vertical|code|cents ), sorted; recompute it from GET /api/v1/verticals or with node tools/matrix-fingerprint.mjs . A test fails if a price changes without a new version and a new row here. Version In force What changed Fingerprint 2026.09.1 12 September 2026 to date First published matrix: four verticals, each a mandatory base layer and four optional modules, 20 prices. Mock rates for validation, not offered to anyone. 06918cd51b6df7c637a972af673957f05a9f7f246aaed5a6f5c49e270aadd600 The first 43 entries in the ledger, recorded on 12 September 2026 before the version field existed, carry no version. They were priced by the same rates: no price has changed since the engine was first deployed that day. Changes What changed, most recent first 17 November 2026 The GitHub Action that seals a release now sits at the root of the public recorder repository, so a workflow uses it as aniljangrabe-dev/riskrouter-recorder ; the older github-action path still works. It now creates its receipts folder itself (a first run on a fresh runner failed before), and the run's summary page lists each file, its log entry and its receipt. A campaign arrival no longer counts on a pointer that only rests on the page: browsers fire a pointer event of their own when a page lays out under it, with nobody there. The pointer must really move, or the visitor scroll, touch or type, as the privacy page says. 28 October 2026 Three new tools in the recorders ( riskrouter-recorder 0.2.0), each keeping records on your side and sending only salted digests: dbseal seals the rows of a PostgreSQL table as they are written, through a trigger you apply yourself; folderseal seals each file that lands in a folder, with receipts that check at /check ; and inspector puts everything sealed in a period into one folder a supervisor checks without us. Recorders Use it from your assistant : Claude, ChatGPT, Gemini CLI, Grok and the OpenAI and Claude APIs, and a page for telecom operators under NIS2. 26 October 2026 Time stamps, checkpoints to witnesses and the outside probe behind /status now run every two hours instead of every hour, so that everything we run on GitHub Actions stays within the plan we pay nothing for. Each new head is still stamped and posted; time not probed is still shown as not measured. 24 October 2026 Checkpoints now go to six witnesses: Geomys's Navigli witness and IPng's krabbel witness join the four of 23 October, each key copied from the operator's own page. MCP clients that cannot send a key, such as Claude's directory connectors, can now sign in to record: the user ticks that entries are permanent and the client receives a free sandbox key, under the same limits as any self-serve key ( how it works ). Reading still needs nothing. A GitHub Action that seals each file of a release and keeps a receipt anyone can verify offline; with an SBOM, each record is a Cyber Resilience Act cra.release . In the public recorder repository, under github-action . A terms of use page for the site, free sandbox keys, the API and the MCP endpoint during validation. 23 October 2026 Two new record templates, for clinical trials ( clinical-trials : the trial master file, its archive, adverse event reports and serious breaches, Regulation (EU) No 536/2014) and for medical devices ( medical-devices : technical documentation and incident reports, the MDR and IVDR), with a page for the software that serves them, /for-life-sciences . Each article was read from its latest consolidated text first. The evidence log is on the public witness network's staging list ( log-list-10qps-4klogs.1 ), and each new checkpoint is now posted to four of its witnesses: GDB Security, Markovian Protocol, Mullvad's staging witness and TrustFabric's staging witness. Each key was copied from that witness's own page; a cosignature is kept only when it verifies, and /witnesses counts them from the files. 20 October 2026 The two hourly jobs, time stamps and checkpoints to witnesses, are now one, so /status is measured once an hour on paper instead of twice; in practice GitHub runs it every few hours, and time not probed is shown as not measured, as before. 19 October 2026 Seal takes a whole folder as well as a file. Every file is fingerprinted in the browser and the folder is one entry in the log, whatever it holds: the receipt lists each file’s path and SHA-256 in the format sha256sum writes, and that list is what is sealed. Checking a folder fails on any file missing or changed and names a file added since; one file from a sealed folder can be checked on its own against the folder’s receipt. The receipt is the same record bundle as before, so both verifiers check it unchanged. 18 October 2026 The evidence log’s newest signed head now goes into Bitcoin every day the log moves, through three OpenTimestamps calendars, over the same digest its RFC 3161 tokens carry. Until now only the quote ledger’s 13 September head was in a block. /age lists every head with its tokens and its block, read from the files when the site is built, and both verifiers check the proofs offline. 17 October 2026 Privacy and the security questionnaire were not complete. They said nobody but the operator sees email sent to us, but the operator reads that mailbox with Claude, an AI assistant made by Anthropic, which drafts replies a person then reads and sends. Both pages now say so, including that Anthropic provides it under consumer terms rather than a data processing agreement. 12 October 2026 Every tool of the hosted MCP endpoint now says what it does to the log: seven read only, and record_evidence appends without changing or removing anything. MCP clients and directories read these hints before letting an agent call a tool. The recorders' source is public at github.com/aniljangrabe-dev/riskrouter-recorder , an exact copy of what we publish to PyPI and npm, with setup steps an assistant can follow for the local MCP server. That server's three tools now carry the same hints: only record_decision writes, and nothing deletes. On npm from the next release. 11 October 2026 Everything you need to check our evidence or build on it is now open source under the Apache License 2.0: the specification and its test vectors, both verifiers, the conformance suite, the regime packs, the integration and verification-page kits and the portable witness, joining the recorders. The public verifier repository carries the licence. The service that runs the log stays private; the design means nobody needs to trust it. A guide for engineers: what the AI Act asks of your logs . Articles 12, 14, 19, 26 and 86 read from the consolidated text that includes the amendment of July 2026, the dates Article 113 now sets for them, what the Act leaves to you, and how to keep logs that hold up, with or without us. The weekly watch now checks every article it cites. The recorders riskrouter-recorder 0.1.1 are on PyPI and npm . The code is unchanged; the README now says the verifiers, the conformance suite and the integration kit are Apache-2.0 too. From this release on, an npm version is staged by the release workflow and made public only when a person approves it with two-factor authentication. 10 October 2026 A weekly watch, with no AI in it, of what the product relies on but does not control: the text of every EU article the regime packs cite, the standards the log implements, and every outside link on this site. Its first comparison found that the SCITT architecture and the COSE hash envelope are now RFC 9943 and RFC 9995, which the specification now cites, and that the MiFID II pack had no record for Article 54(10) of Delegated Regulation (EU) 2017/565 on sustainability preferences: it now has mifid.sustainability-reasons . The AI Act's amendment of July 2026 changed none of the articles the AI Act pack cites. Seal a file takes one click: tick the box that says a seal can never be deleted, press Seal this file , and the page takes a free sandbox key for you, seals the file and checks the receipt on the spot, before you download it. The key stays in the page’s memory for further seals and is shown once; anyone with a key of their own can still paste it. The Node recorder is published as riskrouter-recorder 0.1.0 on npm , Apache-2.0, after the full recorder suite passed: npm install riskrouter-recorder . Its command-line tools run with the package named, npx -p riskrouter-recorder riskrouter-logseal . Recorders 9 October 2026 The Python recorder is published as riskrouter-recorder 0.1.0 on PyPI , Apache-2.0, by the release workflow after the full recorder suite passed: pip install riskrouter-recorder . The Node recorder follows on npm. Recorders 8 October 2026 A roadmap that names what each next step waits on instead of promising dates, a page on what the pilot letter says, and a press page whose figures are computed from the repository when the site is built. The header fits one row on a wide screen, and folds behind a Menu button on a phone; without JavaScript the menu stays open. Every deploy now checks that what answers is what was deployed: the API names the commit, and every page in the sitemap answers with the title the build gave it. Every GitHub Action is pinned to a full commit SHA, every workflow declares the least permission it needs, and a test fails the build if either slips. The site also sends Cross-Origin-Opener-Policy . Security The privacy notice names its controller, Anil Jangra, who builds and runs RiskRouter; the press, continuity and security questionnaire pages say the same. Who is responsible An individual Internet-Draft on digest-only evidence logs is posted at the IETF: draft-jangra-scitt-digest-evidence-00 . No working group has adopted it. Specification 7 October 2026 Completeness chains: records can be numbered in chains, the log refuses a position out of turn with nothing written, and GET /api/v2/evidence/chain signs every position it holds, so a record sealed and later held back shows as a gap. Spot checks draw a sample from a Bitcoin block hash; selective disclosure shows one field of a sealed record; co-sealing checks that two parties signed the same record. New frozen forms, with vectors, checked by both verifiers and the conformance suite. Is this all of it? Regime packs for eleven laws, adding GDPR consent and erasure, consumer credit, anti-money laundering, the Cyber Resilience Act, VAT invoices and public procurement, and new kinds for AI model provenance, screening criteria and content credentials, IDD pre-contract disclosures and product versions, and MiFID II call recordings. Regime packs Invoices The recorders chain records themselves when given a chain secret, fingerprint files kept elsewhere, and seal at the OpenTelemetry Collector through otlp-sealer ; logseal exports a sealed forensic copy. Recorders Deadline proofs compare a sealed report with the NIS2, GDPR and Cyber Resilience Act clocks, and a software vendor can publish a verification page in its own brand that cannot send anything anywhere. Deadline proofs The kit 6 October 2026 Qualified time stamps can now be switched on by contract: the hourly job can use a qualified authority’s credentials and be limited to one token an hour, and it never calls a paid authority without them. Nothing is called qualified until such an authority is listed with its EU Trusted List entry; today none is. FAQ The FAQ and the short map for AI assistants now describe the log as it is: any record a firm may have to defend, built first for software vendors serving regulated finance and for AI platforms. 5 October 2026 The home page now says who we build for first: software for regulated finance and AI-governance platforms , where one integration serves every firm on them, and compares the log with the alternatives , including where they do better. For AI platforms The site is in English only for now. The French, Dutch and German pages are withdrawn, and their old addresses redirect permanently to the English page of the same name. Each article on what the law asks you to record still links to its EUR-Lex text, where every official language is one click away. 4 October 2026 Seal a file , for anyone with no software to integrate: the browser fingerprints the file (it is never uploaded), the salted digest is recorded, and you keep a receipt that the published verifiers check offline. Its check half is at /check. Industries : one page for each kind of record people are later asked to prove, from insurers, lenders and AML teams to hiring, laboratories, sustainability reports, research and public procurement. Each ends with a two-minute test on a real file of your own; no page offers a made-up sample. 30 September 2026 AI agents can connect to the evidence log over the Model Context Protocol at https://api.riskrouter.eu/mcp : read the signed head, proofs, the checkpoint and usage without a key, and record a digest with one. Each tool is an existing route, answered by the same code. There is no pricing tool. How to connect Opening https://api.riskrouter.eu/mcp in a browser now shows a page saying what the address is and how to connect, with its tools; MCP clients get exactly what they did before. And the endpoint is ready to be listed in the official MCP Registry as eu.riskrouter/evidence-log . AI agents and assistants : one page to connect Claude Code, Cursor or any MCP client to the evidence log, with the tools and which need a key, the agent framework adapters and AI Act logging. And /llms-full.txt : every English page as text, for AI assistants, rebuilt with the site. Sealed security logs : logseal , in the recorder kit, seals any log stream segment by segment, so after a breach you can show which logs existed before it and that none was rewritten or removed, while the logs stay on your own storage. A sixth pack, NIS2 and data breaches, records each segment, each incident stage and each report sent, mapped to NIS2 Articles 21 and 23 and GDPR Articles 33 and 34. It detects nothing and names no attacker; it keeps the evidence trustworthy. Load tested on every change to the engine, against the self-hosted stack: the log ceiling (600 entries a minute) and the per-caller limit (60 requests a minute) both held exactly and answered a clean 429, batches of 50 were recorded in about 22 ms, no request failed with a server error, and every accepted entry was in the tree. These are test-machine figures, not production ones. In French, Dutch and German (withdrawn on 5 October 2026): the home page and what the law asks you to record, with each article linked to the EUR-Lex text in that language and a language switch on every page that has a translation. The API, the specification and most pages remain in English, and the translated pages say so. The drop-in recorders are now tested through an outage: the log refusing everything, then writing a batch whose answer is lost, then back, with the process killed in between. Nothing is lost and nothing is marked recorded before the log answered; a lost answer costs a duplicate leaf, never a gap. Recorders Uptime, measured from outside: GitHub Actions, which runs on none of our infrastructure, probes the website and the API on a schedule of twice an hour that GitHub often runs late or skips, in practice every two hours or so, and keeps every answer, and /status shows the last 30 days. Time GitHub did not probe is shown as not measured, never as up; a degraded API is counted apart. The figures as JSON: /uptime.json . Backups are now daily and kept for 90 days, each restored into a fresh database and checked the same day; they were weekly, kept 30 days. At most a day of entries now separates the database from its last backup. Restore drill DORA Article 30, clause by clause : every provision a contract with an ICT provider must contain, the clause we offer for it, and what is in place today, including what is not. Proposed terms until RiskRouter’s legal entity exists. What the law asks you to record : for IDD, MiFID II, the AI Act, GDPR Article 22 and DORA, what each article asks, when the record gets tested, and what a sealed record adds, built from the same files as the regime packs. It says plainly that no law requires RiskRouter or any product. Questions and answers : nineteen short answers to what people ask before they write to us, each linking to the page that holds the detail. Search: a search box in every page’s header, and /search , over every page of the site down to its sections, with exact phrases in quotes and -word to exclude. The index is a file on this site, so what you type never leaves your browser. Press / to jump to it. The mark is now the icon everywhere: browser tabs, phone home screens, search results and any API address opened in a browser. The icon was redrawn so its three squares stay apart at every size, as they do in the header. The API’s own page at api.riskrouter.eu now looks like the rest of the site, with the same mark and icon, and lists every current route. The admin console uses the same icon. 29 September 2026 Agent adapters : each run of an AI agent becomes one evidence record, with its tool calls, model calls, handoffs, guardrails and graph nodes as steps, from a callback handler for LangChain and LangGraph and a tracing processor for the OpenAI Agents SDK, in Python and JavaScript. Steps keep a SHA-256 of the text they handled, not the text. An MCP server gives any MCP client a record_decision tool, whose records say they are the agent’s own statement. Tested against the real frameworks on every change. The recorders and adapters are licensed Apache-2.0 (those packages only; the rest of the source is not licensed), ready for PyPI and npm as riskrouter-recorder . The evidence log’s checkpoint key is published ( anchors/checkpoint/log.vkey ), so /tlog/evidence-v2/checkpoint now serves a signed C2SP checkpoint. Its private half was generated straight into the API’s secret store and exists nowhere else. No witness of the public network cosigns it yet. SCITT : the log registers IETF SCITT Signed Statements that are hash envelopes, signed with a key the distributor registered, and returns RFC 9942 COSE Receipts that any COSE library can check ( POST /api/v2/evidence/statement , GET /api/v2/evidence/receipt ). Only the statement’s digest is kept; the artifact never comes here. Verified in our tests by pycose, an independent implementation, and by both of our verifiers. Nothing frozen changed. Drop-in recorders for Python and Node: a logging handler, a decorator or function wrapper, and OpenTelemetry span processors that turn the decision logs a team already writes into evidence. Each record and its salt stay in a journal on the customer’s side, only the digest is sent, and nothing is marked recorded until the log says so. No dependencies; tested against the real engine and the real OpenTelemetry SDKs on every change. The FSMA replied to our question of 19 September about the boundary of insurance distribution. It restated the legal definitions and left the analysis of whether registration is needed to us, with a specialised legal adviser if need be; it did not rule on our model. Summarised on /evidence , which no longer waits for an answer. 28 September 2026 The evidence log now also speaks the C2SP transparency-log formats that the public witness network uses: a signed checkpoint at /tlog/evidence-v2/checkpoint for the same tree as the v2 head, and hash tiles beside it, so anyone can mirror the tree and compute any proof without our API. Entry bundles are not published. Checked against litewitness, the reference witness, in CI, and by both verifiers. The checkpoint key is not published yet, so no checkpoint is served until it is, and no witness of that network cosigns our log yet. Specification Witnesses : anyone can now run a witness of an evidence log on their own account, as a Cloudflare Worker, in a fork of the verifier repository on GitHub, or on a small server, with a key only they hold. It co-signs a head only after checking the log never rewrote history, pins each log’s keys on first sight, and keeps the evidence for good if a log contradicts itself. Its web side only serves what it has stored. A public registry at /registry.json lists every log and every witness, from files in the repository; a self-hosted instance can witness every listed log, and its own, by switching on one service. Today the registry lists one log, ours, and no witnesses. The engine can run in a customer’s own account: the same code on their own Cloudflare and Supabase, or anywhere containers run (PostgreSQL, PostgREST and the API). A self-hosted instance signs with its own key, holds its own data and calls nothing of ours, and it hands out no sandbox keys unless its operator switches that on. Built and checked end to end from nothing every week in CI. Running it needs the source and a licence, agreed in writing. Continuity 27 September 2026 RFC 3161 time on every signed head: each hour, every new evidence head and ledger attestation is sent to two independent time-stamping authorities (DigiCert and Sectigo), and their signed tokens are kept in anchors/tsa/ beside the Bitcoin anchors. The token binds the exact payload we signed; both verifiers check it offline, and /verify counts only tokens that verify. These are not qualified eIDAS time stamps, and every page and verifier says so. Specification Regime packs : what to record, rule by rule, so the log’s proof answers the question a supervisor will ask. Five packs, for insurance distribution (IDD), MiFID II suitability, the AI Act’s logging and human oversight, GDPR Article 22 and DORA incidents, each field mapped to its article with a link to the text. Published as data at /packs/.json . A record has one canonical form (RFC 8785, specification ), so a supervisor holding the record and its salt can recompute the digest in any language, and a bundle (record, salt, proof) checks offline with either verifier. The conformance suite gains the record cases. Nothing changed in the API: the log still accepts any kind. Signed leaves : a distributor can sign each claim with its own P-256 key, registered once at POST /api/v2/evidence/keys and served to anyone at GET /api/v2/evidence/keys . The signature goes inside the leaf (a new leaf version, v3, beside v2 in the same tree), so a proof now shows that the distributor asserted the decision, not only that we sealed it. Verified before anything is written; a claim that does not verify records nothing. Optional per entry. Both verifiers and the conformance suite check the client signature; the Python kit gains a standard-library signer. A conformance suite for anyone implementing the evidence formats in their own product: it grades an implementation in any language against the specification’s test vectors and against proofs that have been deliberately broken, checks the exports and proofs the product writes, and gives a result file that is the implementer’s own to publish. Ships in the kit at /kit/conformance/ ; the reference code passes all 46 cases. The evidence log’s Merkle tree is now kept in the database and maintained on every append, so a head or a proof reads a handful of stored subtree roots instead of every leaf, and the size ceiling on those routes is gone. Signed heads are stored once per tree size and the database refuses a head whose root is not the tree’s own. Nothing frozen changed: the specification , its vectors and both verifiers are untouched, and the stored-node fold is held equal to the leaf fold by tests in JavaScript and in SQL. POST /api/v2/evidence/batch : up to 500 digests in one transaction, in order, all or nothing, under the same ceilings per entry. API reference . Usage : how much is recorded and by whom, counted live from the ledger by a new public endpoint, GET /api/v1/usage , which returns nothing but numbers and week dates. The figure it leads with is entries recorded by keys that are not ours. Below it, how many independent witnesses and conformance results published by others exist, from the repository at build time; today both say none yet. Every claim, and where to check it : one page for a security review, a regulator or an investor, each claim with its artefact and who other than us can check it, and a plain no where the answer is no. How old the record is : the first anchored head, the earliest Bitcoin block it is in, how many heads are anchored and how many independent witnesses co-sign, computed from the published anchors every time the site is built and never typed into a page. Also on the home page and as proof-of-age.json . Where a figure is zero it says none yet. 24 September 2026 A new look: paper, ink and typefaces served from this site, set like a document rather than a product launch. The home page shows the last three ledger entries exactly as they were anchored. A page for broker-software vendors : what one integration gives every broker on a package. Check your setup : paste a key and see whether it is recording, whether your chain links up and whether we signed its head. Verifiers now choose the signing key by its id, so attestations saved before a future key rotation keep verifying. Every page is checked against WCAG 2.1 AA on every change; code blocks can now be scrolled with the keyboard. A security questionnaire, answered : the questions a vendor-risk team sends, each answer linked to its evidence, with a plain no where the answer is no. A status page that checks the API, the ledger and the evidence log live from your browser, and shows no uptime history until one has been measured. The API names the commit it is running ( build.commit on /api/v1/health ), and a bill of materials shows that no third-party package runs in production. First restore drill: a backup of the ledger was restored and matched the head anchored in Bitcoin. Continuity Backups are now weekly and automatic: each one is restored into a fresh database and checked against the anchored head, then kept encrypted on GitHub for 30 days, which is outside the EU. Sub-processors Fixed: a refused contact-form write logged the database’s whole error body, which can contain the sender’s name and email. Error logs now keep only the error code and message. Fixed: checking an export against a head recorded earlier said history has been altered whenever entries had been added since, which is always. A recorded head now matches when it is any earlier head of the chain, and the check says up to which entry history is unchanged. Same fix on /verify , in tools/verify-ledger.mjs and in the Python verifier. Fixed: when the ledger could not be read, /verify showed 0 entries and its copy button put the error message on the clipboard as if it were a hash. It now shows no count and offers nothing to copy until a real head has arrived. 23 September 2026 Evidence log v2: record a fingerprint of any regulated decision, not only a quote. The record stays with you; only its salted SHA-256 is sent. Entries sit in an RFC 9162 Merkle tree with signed heads, inclusion and consistency proofs. v1 is unchanged. API reference An open specification of every byte the evidence commits to, with test vectors and a second, independent verifier in Python. A witness tool anyone can run to check our heads and co-sign them. There are no independent witnesses yet. How it works An integration kit for broker-software vendors: Python, PHP, C#/.NET and Java, standard library only, plus a reference broker app, all tested against the real Worker on every change. The 13 September head is timestamped in Bitcoin (block 967830 and later), via three OpenTimestamps calendars. Heads are now anchored weekly by an automated workflow. Verify Single-entry proofs: show one quote to a regulator or a customer without revealing any other. Verify Every API error links to the row of the API reference that explains it; ten codes that had no row now do. A Postman collection generated from the OpenAPI spec, and a try it panel on the API reference . Fixed: a browser on riskrouter.eu could not send an API key to the API, so the live console could price but never record. Security headers on the site and the API, a hash-based Content-Security-Policy, security.txt and a disclosure policy . A continuity and exit plan , with the facts a DORA register of information needs. Three guides for distributors, a share image per page, and the ledger’s live head on the home page. Campaign links are counted without cookies, storage or third parties. What is counted 22 September 2026 Self-serve sandbox keys: take one on screen with nobody in the loop. Get a key Idempotency-Key on recorded quotes: a retried request can never create a second ledger entry. An OpenAPI 3.1 description of the API, kept identical to the running routes by a test. Check your own export in the browser on /verify , with nothing uploaded. Fixed: a keyed quote whose ledger write failed still reported attribution.recorded: true . 20 September 2026 Distributors can export their own entries, the chain they sit in and the signed head, and verify it without us. The public verifier repository went live. Heads first submitted to OpenTimestamps calendars. A break it yourself demonstration of the chain on /verify . Rate limit on the export, the one route that had none. 19 September 2026 The platform moved to riskrouter.eu, with a published contact address. 13 September 2026 Attestations of the ledger head are signed (ECDSA P-256), so a head someone saved cannot be disowned. Ledger writes are attributed to a distributor, and the attribution is inside the entry’s hash. 12 September 2026 The ledger becomes tamper-evident: every entry is SHA-256 hash-chained to the one before it. First version of the engine, the audit ledger and the demo console. Rating matrix 2026.09.1 . ## Open evidence specification URL: https://riskrouter.eu/spec Specification version 1.0 23 September 2026 Open evidence specification Every byte RiskRouter’s evidence commits to, written down so that anyone can check it without us, and anyone can build a second implementation. Two already exist that share no code: the JavaScript verifier and a Python one, held to identical verdicts by the repository’s tests. Anyone may implement this specification without asking us: the specification, its test vectors, both verifiers and the conformance suite are published under the Apache License 2.0 . The completeness, disclosure and sampling constructions, and the SCITT profile, are also written up for the IETF as an individual Internet-Draft, draft-jangra-scitt-digest-evidence , posted on 8 October 2026. No working group has adopted it. Where the draft and this page differ, this page and spec-vectors.json are what was published. 0 Conventions Conventions Hashes are SHA-256, written as 64 lowercase hexadecimal characters. Strings are UTF-8. Fields are joined with | , with no spaces and no escaping; no field may contain | . Timestamps inside hashed strings are UTC, YYYY-MM-DDTHH:MM:SS.ffffffZ , six fraction digits. A reader that receives +00:00 , a space instead of T , or fewer digits must normalise to that form first. Signatures are ECDSA over P-256 with SHA-256, encoded as the 64-byte r||s (IEEE P1363), base64. Public keys are JWK. The published key is in the public verifier repository , key id dd4b4b394c95586a , and stays there after any rotation. Frozen means: changing it would invalidate evidence people already hold, so it will never change within a version. 1 v1 quote ledger The quote ledger: a hash chain 1.1 Canonical form (frozen) chain_index|id|created_at|active_vertical|components|premium|compliance_status|matrix_version|distributor_id|prev_hash components : every component key in ascending order, as key=true or key=false , joined with commas. A value is true only if it is the JSON literal true . premium : exactly two decimals, such as 2.00 . matrix_version and distributor_id : empty when absent. row_hash = SHA-256 of the canonical form. The first entry’s prev_hash is 64 zeros; every other entry’s is the previous entry’s row_hash . The last row_hash is the head . 1.2 Attestation (frozen payload) riskrouter-ledger-attestation|v1|as_of|entries|head_hash The signed statement that the ledger had entries entries and that head at as_of . A verifier rebuilds this payload from the attestation’s own fields; it never trusts a signed_payload field alone. 1.3 Export and single-entry proof An export ( riskrouter-ledger-export|v1 ) carries the holder’s entries in full, the (chain_index, row_hash, prev_hash) of every entry, and the signed attestation. A single-entry proof ( riskrouter-entry-proof|v1 ) carries one entry, the links from it to the head, and the attestation. What these prove, precisely: the holder’s entries are rebuilt from content; other entries are followed by digest only and bind because anyone recomputing the full ledger must reach the signed head. A rebuilt ledger is caught by comparing with a head recorded earlier, which is why heads are anchored (section 4). 2 v2 evidence log The evidence log: digests of any regulated decision, in a Merkle tree 2.1 What the client sends The client hashes its own record and sends only record_digest and a kind tag ( ^[a-z0-9][a-z0-9.-]{0,39}$ ). The record never leaves the client. The client must salt the record. Hash a random value of at least 128 bits together with the record, and keep both. A digest of a guessable record, such as approved , can be confirmed by anyone who guesses it. The salted construction this specification recommends is SHA-256(salt || record) , but any construction the client can reproduce is valid: the log commits to the digest, not to how it was made. 2.1b Canonical record and bundle (frozen) How the record is hashed is the client’s choice, but a record another party must be able to re-hash, such as one written to a regime pack , uses this construction: record = a JSON object; keys ^[a-z][a-z0-9_]{0,63}$ at every level; values are strings (valid Unicode), integers within (2^53 1), true, false, null, arrays and objects, nested at most 32 deep; no fractions (money is integer cents) canonical = RFC 8785 (JSON Canonicalization Scheme) of the record record_digest = SHA-256(salt || UTF-8(canonical)), salt 16 to 64 random bytes as lowercase hex bundle = { "format": "riskrouter-record-bundle|1", "record", "salt_hex", "proof" } A record outside the profile is refused rather than coerced. A bundle is intact when the record and salt produce the entry’s record_digest , the record’s kind is the entry’s, and the proof verifies as below. The vectors carry records chosen for the places two languages differ (escapes, astral characters, key order, edge integers) and records that must be refused. 2.2 Leaf (frozen) leaf string = riskrouter-evidence-leaf|v2|leaf_index|created_at|distributor_id|kind|record_digest leaf hash = SHA-256(0x00 || leaf string) leaf_index counts from 0 with no gaps. created_at is set by the log, never by the client. Every entry carries leaf_version , 2 or 3; a verifier rebuilds the leaf string for the version it sees. 2.2b Signed leaf (frozen, v3) A v2 leaf proves that the log sealed a digest at a time. It does not prove that the distributor asserted anything: for that, the distributor signs its own claim, and the signature goes inside the leaf, so it cannot be swapped after the fact. Optional per entry; an unsigned entry is a v2 leaf exactly as above. Both versions sit in the same tree under the same head. claim payload = riskrouter-evidence-claim|v3|kind|record_digest|signer_key_id|claimed_at signed by the client: ECDSA P-256 over SHA-256, IEEE P1363 (r || s), base64 leaf string = riskrouter-evidence-leaf|v3|leaf_index|created_at|distributor_id|kind|record_digest|signer_key_id|claimed_at|client_signature leaf hash = SHA-256(0x00 || leaf string) signer_key_id = first 16 hex characters of SHA-256(JSON [x, y] of the public JWK), as a witness id is derived claimed_at = the client's own timestamp, YYYY-MM-DDTHH:MM:SS.ffffffZ, exactly as it was signed The client signs only what it knows before sending; leaf_index and created_at remain the log’s. The log verifies the signature before recording and refuses a claim that does not verify, with nothing written; the database then commits to the signature as presented, and every verifier checks it again. The public key is registered once ( POST /api/v2/evidence/keys ) and served to anyone ( GET /api/v2/evidence/keys?key_id= ); a proof of a v3 leaf carries it as signer_public_key . The signer should publish the same key under its own control , exactly as a witness does, so that the client signed it is not the log’s word alone. A key id belongs to one distributor forever; registering a new key never removes an old one, so leaves signed before a rotation stay verifiable. What a v3 leaf proves: this distributor, holding this key, asserted this digest and this kind at the time it claimed, and the log sealed that assertion at created_at in a tree whose head is signed. What it does not prove: that the record behind the digest is true, or that the key was not stolen. 2.3 Tree RFC 6962 / RFC 9162, unchanged: node hash = SHA-256( 0x01 || left || right), with the split at the largest power of two smaller than the number of leaves; the root of the empty tree is SHA-256 of the empty string. Inclusion and consistency proofs are the RFC 9162 section 2.1.3 and 2.1.4 constructions, verified by the algorithms given there. The implementation is checked against RFC 6962’s own reference roots. 2.4 Signed tree head (frozen payload) riskrouter-evidence-head|v2|tree_size|root_hash|timestamp Signed with the same key as v1 attestations. A verifier must take tree_size and root_hash from the same signed head : an inclusion proof does not bind the tree size on its own. A head with no key bound is returned unsigned and says so; it is never given a signature it does not have. 2.5 Endpoints POST /api/v2/evidence , POST /api/v2/evidence/batch , GET /api/v2/evidence/head , GET /api/v2/evidence/proof , GET /api/v2/evidence/consistency , GET /api/v2/evidence/chain ( 2b ), and for signed leaves POST and GET /api/v2/evidence/keys , described in the API reference and openapi.json . A consistency proof is checked against the root of a head the verifier saved, never a root the log supplies. 2b Completeness Chains, spot checks, disclosure and co-sealing A proof of one record answers was this changed? . An auditor’s harder questions are is this all of it? , did you keep it? , what did this one field say? and did both of you agree to it? . The four constructions below answer them. Each is optional per record, none changes a leaf, and all four are frozen from their first published vectors (settled 7 October 2026). 2b.1 Completeness chains (frozen) A firm puts records in numbered chains. The record carries three fields inside itself, and the entry carries two of them beside the digest, outside the leaf: record fields chain_tag 64 hex characters; recommended: HMAC-SHA256(secret, "riskrouter-chain-tag|1|" || name) chain_seq 1, 2, 3, ... with no gap chain_prev the record_digest of the record at chain_seq - 1; 64 zeros at position 1 entry fields chain_tag, chain_seq sent with record_digest; never part of the leaf string link line chain_seq|leaf_index|created_at|record_digest\n links_digest SHA-256(UTF-8(the link lines from `from` to `to`, concatenated)) chain payload riskrouter-evidence-chain|v1|chain_tag|length|tree_size|root_hash|timestamp|from|to|links_digest The log refuses a chain_seq that is not the next position, and a chain_tag another distributor uses, with nothing written. GET /api/v2/evidence/chain?chain_tag= answers what the log holds in that chain as of the current signed head: its length and, for each position, the leaf, the time and the digest, signed over the chain payload with the head-signing key. A holder checks the signature, recomputes links_digest , and then that each record shown produces the digest at its position, names that position, and points at the one before. A position the statement lists and nobody showed is a record left out. A firm alone cannot hide one; doing so would need the log to sign a statement its own tree contradicts. On a log the firm runs itself, this guarantee is the firm’s own word. completeness bundle = { "format": "riskrouter-completeness-bundle|1", "chain": [statement, ...], "records": [{ record, salt_hex }, ...], "as_of"? } With as_of , only the positions sealed by that time must be shown. 2b.2 Spot checks (frozen) An auditor announces, and seals in the log, a future Bitcoin block height and a sample size. When the block is mined its hash is the seed, and neither side could have chosen it: for counter = 0, 1, 2, ... h = SHA-256(UTF-8("riskrouter-spot-check|v1|" seed "|" population "|" counter)) x = first 8 bytes of h, big-endian unsigned skip if x >= 2^64 - (2^64 mod population) position = 1 + (x mod population), kept unless already drawn until min(sample, population) positions are drawn spot-check bundle = { "format": "riskrouter-spot-check-bundle|1", "chain", "seed", "bitcoin_height", "sample_size", "records" } population is the chain’s length in the signed statement. The firm shows the records at the drawn positions; nobody sends anyone a bulk copy. The verifier recomputes the draw; that the block at that height has that hash, and was mined after the plan was sealed, the auditor checks on any Bitcoin node. 2b.3 Selective disclosure (frozen) field salt HMAC-SHA256(secret, UTF-8("riskrouter-disclosable|1|" || name)) 32 bytes, one per top-level field commitment SHA-256(field salt || UTF-8(canonical({name: value}))) sealed record { "format": "riskrouter-disclosable|1", "kind": , "fields": { name: commitment, ... } } record_digest SHA-256(salt || UTF-8(canonical(sealed record))) as for any record (2.1b) disclosure { "format": "riskrouter-disclosure-bundle|1", "record": sealed record, "salt_hex", "proof", "disclosed": { name: { "value", "field_salt_hex" }, ... } } The secret never leaves the firm: a disclosure carries the salts of the fields it shows and nothing that derives the others, so a hidden low-entropy value (a score, an outcome) cannot be guessed. Field names are visible; values are not. 2b.4 Co-sealing Two parties who share a record and its salt each record the same digest as a signed (v3) entry, under their own key and their own account. A co-seal ( riskrouter-coseal|1 : bundles , optionally expected_signers , the keys each party published) holds when every bundle verifies, the digests are equal, every entry is v3, and the signers and the distributors are all different. Nothing new is frozen: it is a rule over leaves that already exist. Both verifiers and the conformance suite check all four; spec-vectors.json carries a chain statement and its payload, four spot-check draws (one seeded with Bitcoin’s genesis block hash), and a disclosure with its field salts, commitments and sealed record. Design notes: docs/completeness.md , docs/spot-checks.md , docs/selective-disclosure.md , docs/co-sealing.md . 3 Witnesses Independent witnesses A witness is anyone who is not the operator. Each time it runs it fetches the current signed head, checks the operator’s signature, requests a consistency proof from the last head it saved, and checks that proof against the root it saved. Only if every check passes does it co-sign: riskrouter-evidence-cosign|v2|witness_id|tree_size|root_hash|cosigned_at If a check fails, the witness holds two heads, both signed by the operator, that cannot both be true. That pair is evidence of a rewritten history that the operator cannot disown, produced by someone the operator does not control. The reference witness is tools/witness.mjs ; it saves the pair and exits with status 2. The portable witness in witness/ of the verifier repository does the same for many logs at once, on Cloudflare Workers, Node or GitHub Actions. Its cosignature file is the same format, with one added member, log ( { id, api } ), which a verifier ignores. It pins a log’s head-signing keys the first time it sees them and raises an alarm if a later list drops or changes one; an alarm is never overwritten, and the witness stops co-signing that log. It serves, read-only, /public-key.json , /logs.json , /cosignatures//latest.json and /alarms/.json . Logs and witnesses are listed at /registry.json ( riskrouter-witness-registry|1 ); a witness is listed only once it publishes its key under its own control. There are no independent witnesses yet running this tool; six staging witnesses of the public witness network cosign the same tree's C2SP checkpoints ( below ). How to become one: /witnesses . The same log in the C2SP formats The v2 tree is also published in the formats the public witness network speaks, so that witnesses who have never run our code can check it: c2sp.org/signed-note , tlog-checkpoint , tlog-cosignature and tlog-tiles . Two things are frozen, on the same terms as the forms above: the origin, api.riskrouter.eu/tlog/evidence-v2 , and the checkpoint text, three lines and no extension lines: api.riskrouter.eu/tlog/evidence-v2 The root is the same 32 bytes as root_hash in the v2 head, in base64 instead of hex, and one read of the tree gives both. The note is signed with the log’s Ed25519 key (signature type 0x01 ), whose key name is the origin and whose key ID is the first four bytes of SHA-256(name || 0x0A || 0x01 || public key). That is a separate key from the P-256 head-signing key, published as a vkey in anchors/checkpoint/log.vkey . A checkpoint is served at GET /tlog/evidence-v2/checkpoint only when that key is bound; otherwise the answer is 503 , never an unsigned note. Witnesses of the network return cosignatures of type 0x04 (cosignature/v1, Ed25519 over cosignature/v1\ntime \n and the checkpoint text), and we keep each one that verifies in anchors/checkpoint/cosigned/ . Tiles of hashes are served at /tlog/evidence-v2/tile//[.p/] , so anyone can mirror the whole tree and compute any proof without our API ( node tools/checkpoint.mjs mirror ). Entry bundles are not published: every leaf names a distributor and a kind, and a distributor holds its own entries with their proofs. Both verifiers check checkpoints and cosignatures; spec-vectors.json carries a checkpoint and a cosignature made with the RFC 8032 test keys, and the conformance suite derives cases from them. 3b SCITT Signed Statements and COSE Receipts The log also serves as a Transparency Service in the sense of the IETF SCITT architecture (RFC 9943), returning receipts as defined by RFC 9942 (COSE Receipts). Registration, POST /api/v2/evidence/statement , applies the policy riskrouter-scitt-profile|1 : a COSE_Sign1 signed ES256 with a key the caller registered ( kid = its 16-hex id as UTF-8), CWT claims iss and sub (text, neither an email address), and a hash envelope (RFC 9995: protected 258 = -16, a 32-byte payload), so the artifact never reaches the log. What the log keeps is one v2 leaf of kind scitt.statement whose record_digest is: SHA-256( 0xD2 0x84 || bstr(protected) || 0xA0 || bstr(payload) || bstr(signature) ) that is, the statement re-encoded with an empty unprotected header. A receipt ( riskrouter-scitt-receipt|1 ) is a COSE_Sign1 signed ES256 with the head-signing key ( kid = its key id), vds 395 = 1 (RFC9162_SHA256), vdp 396 holding one inclusion proof [tree_size, leaf_index, [hash, ]] at label -1, and a detached payload: the root. Its protected header carries CWT claims ( iss , sub = the statement digest, iat ) and riskrouter-leaf = [leaf_index, created_at, distributor_id, kind] . To check one: rebuild the v2 leaf string from those fields and sub , walk the proof to a root, and verify the signature over ["Signature1", protected, h'', root] with our published key. Both verifiers do this, and pycose, an independent COSE library, verifies our receipts in our tests. The statement itself stays with its issuer: we keep its digest only, which is where we depart from the architecture draft, deliberately (rule 15). spec-vectors.json carries a statement digest and a receipt’s signed bytes. 4 Anchoring Anchoring heads in time Signed v1 heads are committed to OpenTimestamps calendars and upgraded to Bitcoin attestations; the 13 September 2026 head is in block 967830 and later. Since 18 October 2026 the newest v2 head of each day is anchored the same way: its payload alone, UTF-8 with no trailing newline, is kept as anchors/ots/evidence-head-N.payload , and the proofs beside it commit to its SHA-256, the same imprint as 4.1; ots verify checks one with no code of ours, and /age lists every head with its block. This bounds when a head existed, from above. It says nothing about the moment of each entry before it. 4.1 RFC 3161 time-stamp tokens (frozen imprint) Every two hours, each new signed head (v2) and each new signed attestation (v1) is sent to the RFC 3161 time-stamping authorities listed in anchors/tsa/authorities.json . The token binds exactly the payload we signed: message imprint = SHA-256(UTF-8(signed payload)) hashAlgorithm sha256 signed payload = riskrouter-evidence-head|v2|tree_size|root_hash|timestamp or riskrouter-ledger-attestation|v1|as_of|entries|head_hash file = anchors/tsa/..json { "format": "riskrouter-head-timestamp|1", subject, signed, signature, payload, tsa: { id, name, url, qualified, trusted_list }, token (DER, base64), gen_time, requested_at } A verifier rebuilds the payload from the head in the file, checks our signature on it, then the token: the imprint, the signed attributes (content type and message digest), the authority’s signature with the certificate the token carries, that the certificate is for time-stamping and valid at genTime , and the chain the token carries, whose root it compares with the root the authority publishes ( openssl ts -verify -data payload -in token -token_in -CAfile root.pem does the same). Both verifiers do this offline. None of the authorities used today is a qualified eIDAS time-stamping service : a token is evidence of time from an independent authority, without the presumption of accuracy that a qualified time stamp carries under Article 41(2) of Regulation (EU) No 910/2014. A file calls a token qualified only when its authority is listed with its entry on the EU Trusted List. 5 Test vectors and implementations Check your implementation spec-vectors.json holds deterministic inputs and expected outputs for every frozen form above: timestamp normalisation, a v1 canonical form and row hash, the attestation payload, eight v2 leaves and the roots of every prefix, an inclusion proof, a consistency proof, a head payload, a cosignature payload, a C2SP checkpoint and cosignature with their Ed25519 signatures, a chain statement, spot-check draws and a selective disclosure. It is regenerated from the reference code and both verifiers must agree with it before any change ships. JavaScript : tools/verify-ledger.mjs , tools/merkle.mjs , tools/witness.mjs (Node, standard library). Python : riskrouter_verify.py (standard library only, including its own P-256 and Ed25519 checks). In the browser : the checker on /verify , which sends nothing anywhere. Run the conformance suite The conformance suite turns those vectors into cases an implementation in any language can answer, adds proofs that have been deliberately broken and must be refused, and grades the answers. It ships with the reference code and the vectors in /kit/conformance/ , needs only Node, and contacts nothing. node conformance.mjs cases > cases.json # one JSON object per frozen form, with the expected answer # your implementation answers each case: { "": } node conformance.mjs grade answers.json --name "Your product 12.4" > result.json # or drive a command that reads one case on stdin and prints its answer: node conformance.mjs run --command "python3 my_impl.py" --name "Your product 12.4" > result.json # and check what your product produced: an export, an entry proof, an evidence proof, an evidence pack node conformance.mjs artefacts ./evidence/ --key signing-key.json > artefacts.json The result file is the implementer’s own statement: that this implementation, on this date, answered these cases as the vectors say, named by the vector set ( riskrouter-spec-vectors|1 ) and the digest of the cases, so anyone can run the same cases again. RiskRouter certifies nothing and nobody. An implementation that passes needs no permission from us to say so, and no relationship with us to keep it true: the vectors and the suite are published, and both are free. A defect in this specification, or a disagreement between it and any implementation, is a bug we want to hear about: security explains how to report one. ## Service status · RiskRouter URL: https://riskrouter.eu/status Status checked live, from your browser Service status Each line below is checked now, by your browser, against the live API. Nothing on this page is typed in by us, so it cannot say all systems operational while something is down. Right now What answered, and what it said Check Result Detail API checking Quote ledger reachable from the API checking Signed ledger head checking Signed evidence-log head checking Running build checking Check again What degraded means here. If the ledger cannot be reached, quotes are still priced but nothing is recorded, and every response says recorded: false . We never report a record that was not written. That is the designed safe state, not a hidden one. History Measured from outside, 18 October 2026 to 17 November 2026 GitHub Actions, which runs on none of our infrastructure, probes the website and the API on a schedule of once every two hours that GitHub often runs late or skips, so in practice every few hours, and keeps every answer. The API counts as up only when it answers OPERATIONAL ; degraded, meaning priced but not recorded, is counted separately. These are the last 30 days, as of when this page was built (2026-11-17 17:04 UTC). Check Up Rounds up Degraded Down API and ledger 100.00% 118 of 118 Website 100.00% 118 of 118 118 rounds. For 704 hours in 103 stretches GitHub ran no probe; that time is not measured, and not counted as up. No round failed. The same figures, as JSON: /uptime.json . Incidents and anything that affected recorded evidence are written up on the changelog . How we would recover from losing the database, and when we last proved it by restoring a backup, is on the continuity page . Dependencies What we depend on, and their own status pages Provider What stops if it stops Their status Cloudflare The site and the API cloudflarestatus.com Supabase (Frankfurt) Recording; pricing continues and says so status.supabase.com Evidence you have already saved does not depend on either: an export, a proof or an attestation verifies offline, with no call to us. How to check one . ## Security questionnaire, answered URL: https://riskrouter.eu/security-questionnaire For vendor-risk and security teams answers as of 23 September 2026 Security questionnaire, answered The questions a bank’s or insurer’s vendor-risk team sends, answered before you send them. Every answer links to something you can check: a page, a live endpoint, a published file, or a test in our repository. Where the answer is no, it says no. If your own questionnaire asks something that is not here, send it to anil@riskrouter.eu and the answer will be added here for the next team too. Read this first Three facts that frame every answer RiskRouter is a validation build operated by one person, Anil Jangra. There is no legal entity yet, no staff, and no contract yet. Answers below describe what is in place now, not what a larger company would have. It holds almost nothing worth stealing. Quotes carry no personal data; the v2 evidence log holds only SHA-256 digests; API keys are stored only as digests. The one table with personal data is contact-form enquiries, deleted after twelve months. Your evidence does not depend on our security. A ledger export or proof you hold verifies offline, against a key we published and heads anchored in Bitcoin. If we were breached, a rewrite of history would be detectable by you, without us. How . Evidence marked repository lives in our source repository, which is private today. Any file named here is shown to a prospective customer on request. 1 Data What data, where, for how long Question Answer Evidence Do you process personal data on our behalf? No, by design. A quote is a vertical and a set of true/false options. Anything else in a request is dropped before storage. The evidence log accepts only a 64-character digest and a short tag. /trust ; repository: tests/pii.test.mjs sends personal data in every field and asserts none is stored What personal data do you hold at all? Contact-form enquiries (name, email, message), and the administrator’s own email address in the admin audit log. /privacy Retention Enquiries are deleted after 12 months by a scheduled database job, running daily. Ledger and evidence entries are kept permanently; that is their purpose, and they hold no personal data. /privacy ; repository: schema.sql , purge_expired_contact_enquiries() Where is data stored? Supabase, eu-central-1 , Frankfurt, Germany. /trust , /continuity Where is it processed? Not only in the EU. Requests are handled by Cloudflare at the location nearest the caller, which may be outside the EEA. Nothing is stored there. EU-only processing would need Cloudflare’s regional services, which we do not have. /trust , /privacy Encryption in transit Yes. HTTPS only, with HSTS on the site. The Workers reach the database over HTTPS. Response headers of any page ( Strict-Transport-Security ) Encryption at rest Provided by the database host, Supabase, under its own security documentation. We add no application-level encryption, and have not independently verified theirs. Supabase’s documentation Can you delete our data on request? Enquiries: yes. Ledger and evidence entries: no, never , because an audit trail that can be deleted is not one. That is why they may hold no personal data, and why a self-serve key must acknowledge it before recording. /privacy , /integrate 2 Access control Who can reach what Question Answer Evidence Who has administrative access? One person, the operator. No one else holds production access. /continuity How is the admin console protected? Cloudflare Access in front, restricted to one email address; the Worker then verifies the Access signature itself rather than trusting a header, and checks the address against its own list. Writes are accepted only from the console’s own origin. Repository: admin-worker.js , tests/admin.test.mjs Is multi-factor authentication enforced? Not claimed. The minimum configuration is a one-time code sent to the allowed address. Stronger factors depend on the identity provider linked to Access, and we do not claim one here. — What can an administrator do? Five things and no more: create a distributor, suspend or reactivate one, issue a key, revoke a key, set an enquiry’s status. There is no path from the console to a quote, a premium, a ledger entry or the schema. Every action is one database transaction that also writes an append-only audit row. Repository: CLAUDE.md rule 8, schema.sql , tests/admin.test.mjs How are your customers authenticated? A per-distributor API key, shown once, stored only as its SHA-256, revocable immediately. An unknown, revoked or suspended key is refused with 401 and nothing is written. /docs ; repository: tests/worker.test.mjs Can one customer read another’s records? No. An export carries the caller’s own entries in full and only the hashes of anyone else’s, which is what lets the chain be checked; an evidence proof for another distributor’s entry answers exactly like a missing one. Row-level security is forced on every table and the public roles have no grants. Repository: tests/evidence.test.mjs , tests/schema.test.sql Single sign-on for our staff? No, and there is nothing to sign in to. Customers integrate with keys; there is no customer console. — 3 Integrity Can records be altered? Question Answer Evidence Can a record be edited or deleted? No. Database triggers refuse UPDATE and DELETE on the ledger, the evidence log and the admin audit log, for every role including our own service role; the evidence log also refuses TRUNCATE by trigger. Repository: schema.sql , tests/schema.test.sql Would a rewrite of history by you be detected? Yes, by you. Entries are hash-chained (v1) or Merkle-logged (v2); heads are signed with a published key and anchored in Bitcoin through OpenTimestamps. A rebuilt history produces a different head than one you or a witness saved. /spec , /verify Can we verify without you? Yes. Two independent verifiers, JavaScript and Python, standard library only, plus an in-browser checker that sends nothing. /spec , Python verifier Do independent parties co-sign? Yes, since 23 October 2026. Six staging witnesses of the public witness network (GDB Security, Geomys, IPng Networks, Markovian Protocol, Mullvad VPN and Google’s TrustFabric team) cosign each new C2SP checkpoint; each cosignature is kept only after it verifies. They run on a best-effort basis. Nobody runs our own portable witness yet. /witnesses 4 Development and change How the code is built and shipped Question Answer Evidence How do changes reach production? Every push to main runs the unit, schema, browser and integration-kit test suites; the deploy job runs only after all of them pass, and then checks that the live API names the commit it deployed and that every page answers as built. The running commit is published, so you can see whether production came from that path. /status , GET /api/v1/health ( build.commit ) Is every change reviewed by a second person? No. There is one developer. Review is by automated tests, several of which enforce the security rules mechanically (for example, that each Worker can reach only its named destinations). Repository: tests/isolation.test.mjs Third-party code in production None. The Workers import only our own files; the build fails if that ever changes. Build and test tools are listed separately and never deployed. /sbom.json (CycloneDX) Dependency updates Dependabot watches the build tools and CI actions weekly. Every CI action is pinned to a full commit SHA and the deploy tool to one version; a test fails if either is not. Repository: .github/dependabot.yml , tests/workflows.test.mjs Build pipeline permissions Each workflow declares the least permission its token needs; the workflow that tests and deploys can only read the repository. No workflow runs code from a fork with secrets, and none prints a secret. Repository: tests/workflows.test.mjs Web application protections A hash-based Content-Security-Policy on every page (no inline script runs unless its hash is listed), HSTS, frame denial, no third-party script except Google’s analytics tag and only after the visitor accepts, per-address rate limits on the API, and database-enforced write ceilings that hold even if the API is bypassed. /security ; response headers Secrets management Credentials live only in Cloudflare Worker secrets and GitHub Actions secrets. None is in source code; tests fail if a Supabase URL or key appears in a Worker, or reaches a response. Repository: tests/isolation.test.mjs , tests/worker.test.mjs Penetration test None. If you need one before a pilot, it is a reasonable condition to agree. /security API change policy The formats evidence depends on are frozen permanently. Every change is listed in the changelog, and the contract terms we offer give at least 90 days’ written notice before we remove or change anything you rely on. /spec , /changelog , /dora 5 Logging, incidents, continuity When something goes wrong Question Answer Evidence What is logged? Cloudflare’s request logs for the Workers, and our own error lines when a write is refused: the HTTP status and the database’s error code and message, never a credential and never the rejected row. (Checking this answer found that a refused contact-form write logged the database’s whole error body, which can contain the sender’s name and email. Fixed on 23 September 2026, with a test.) Every administrative action is in an append-only audit table. Repository: worker.js , tests/pii.test.mjs , schema.sql ( admin_events ) Monitoring and alerting Uptime measured from outside since 30 September 2026: GitHub Actions, which runs on none of our infrastructure, probes the site and the API on a schedule of once every two hours that GitHub often runs late or skips, in practice every few hours, and /status shows the last 30 days, with time not probed shown as not measured. The API reports its own state honestly (it probes the database rather than trusting its configuration). No paging service and no 24/7 on-call. /status , /uptime.json ; repository: uptime/ What happens when the database is unavailable? Quotes are still priced; nothing is recorded, and every response says recorded: false . A failed write is never reported as a success. /docs ; repository: tests/worker.test.mjs Incident response Written runbooks for incidents, leaked credentials (each secret, with what its holder could and could not do), key rotation and restore. Our runbook sets notification of affected distributors within 24 hours of a critical incident; that becomes a contractual commitment only in a contract. Repository: docs/runbooks/ (shown on request) Backups No managed backups. The database runs on a plan that keeps none. We take our own every day, automatically, and keep each for 90 days on GitHub, encrypted (AES-256) before upload with a passphrase GitHub does not hold. /continuity ; repository: .github/workflows/backup.yml Has a restore been tested? Yes, every day, automatically, since 23 September 2026; every run so far has passed. All entries restored; the restored head matched the head anchored in Bitcoin; the restored ledger still refused edits and deletions. /continuity ; repository: docs/drills/ , tests/restore-drill.test.mjs Recovery time and point objectives (RTO / RPO) None committed. Loading a backup takes under a second at today’s size; recovery time is dominated by detection and repointing. Data recorded after the last backup would be lost with the database. /continuity Exit plan if you stop Documented: what you keep, what stops, the steps to leave, and facts for a DORA register of information. /continuity Capacity, measured Load tested on every change against the self-hosted stack, not production: 600 evidence entries a minute across the log (864 000 a day) and 60 requests a minute per caller address, both held exactly and refused with a clean 429; batches of 50 recorded in about 22 ms; no server error, and every accepted entry was a leaf. Repository: docs/load-test.md , tools/loadtest.mjs DORA Article 30 contract terms Each required provision, the clause we offer, and what is in place today. Proposed until a legal entity exists. /dora 6 Third parties Sub-processors and suppliers Question Answer Evidence Sub-processors Cloudflare (runs the site and API), Supabase (database, Frankfurt), GitHub (encrypted daily backups, outside the EU), Microsoft (the mailbox behind our published address), Anthropic (Claude, which the operator uses to read that mailbox and draft replies, and to build and run the service; under its consumer terms, not a data processing agreement). That is the complete list. /privacy What can the API contact? Exactly one destination: the database. It contacts no carrier, bank, payment processor, analytics or advertising service. A test fails if a new destination is added. /trust ; repository: tests/isolation.test.mjs Certifications None. No ISO 27001, no SOC 2. Our providers publish their own; that does not make us certified, and we do not say it does. /security Insurance (professional indemnity, cyber) None yet. — Regulatory status Not a registered insurance intermediary and not an insurer. RiskRouter is software for licensed distributors and sells no cover. /compliance Report something Found a problem, or an answer that is wrong? Security reports: /security and security.txt . An answer on this page that is no longer true is a bug of the same kind; tell us and it will be corrected here and on the changelog . ## For software vendors serving regulated finance · RiskRouter URL: https://riskrouter.eu/for-software-vendors For the companies that build software for insurers, brokers, investment advisers and lenders For software vendors serving regulated finance Your customers are asked, sometimes years later, what they advised or decided and why: a demands-and-needs note, a suitability statement, a credit assessment. They need to show the record and show it has not been changed since. You integrate RiskRouter once, and every firm on your software can do both. The file stays in your product; only a salted fingerprint of it reaches us. What changes for your customers What a firm on your software gets A fixed record of each advice or decision When the record is saved, your software sends a salted SHA-256 of it to the evidence log. From that moment it cannot be altered without the fingerprint showing it. The record, and the client’s name, stay in your database. Insurance distribution Investment advice Credit A date nobody can move Every entry sits in a Merkle tree whose head we sign, time-stamp with two independent authorities (not qualified under eIDAS) and anchor in Bitcoin. Neither your customer, nor you, nor we can back-date a record. An evidence pack per client For one client: the record, its salt and the proof. An inspector, an ombudsman or a court checks it offline with public tools, and learns nothing about any other client. See the reference app that builds one . The work involved What the integration involves Take a sandbox key. No form and no contract: one click on /integrate . Run the example in your language. PHP, C#/.NET, Python and Java, each using only the language’s standard library: a client of 60 to 90 lines and a short demo. Each one records a note’s fingerprint, fetches its proof and verifies it offline. The integration kit . Wire two calls into your product. POST /api/v2/evidence with the record’s fingerprint when it is saved, and GET /api/v2/evidence/proof when a customer needs the evidence. Up to 500 records go in one call with POST /api/v2/evidence/batch . Keep the salt with the note , in your own database. Without it nobody, including us, can link a fingerprint back to a person. The reference broker desk does all of it in one Python file of about 150 lines. Every example in the kit is run against the real API code on every change we make, so an example that stopped working would fail our own build. The pricing example in the documentation is a worked example, not part of the offer: you record fingerprints of the records your product already produces. The format is published , so you may also implement it yourself without asking us, and check that implementation with the conformance suite : it grades your code against the specification’s own test vectors and checks the exports and proofs your product writes, offline, and gives you a result file that is yours to publish. Nothing about it depends on us. What we do not do We will not compete with you or your customers RiskRouter never contacts the customer, never handles a claim, never holds the carrier relationship and never stores the customer file. Those are fixed boundaries of the product, not a roadmap we have yet to reach. Your customers stay yours. Working together What a first integrating vendor can expect No vendor has integrated yet, so there is no programme to join and no logo wall to be added to. Here is what we can promise today: Your developers talk directly to the person who builds the API. What you need goes to the top of what gets built next, if it fits the rules the product keeps. Your name appears on this site only once you have agreed to it in writing. No exclusivity and no lock-in: the specification is open and the verifiers are free. The sandbox is free. There are no commercial terms yet; we would agree them with you, not announce them to you. A pilot is recorded in a one-page letter of intent: no fees, no exclusivity, no personal data, and either side can end it at any time. What the letter says What comes next, and what it waits on Before you commit For your security and compliance review The security questionnaire is answered in advance, with evidence and plain answers where the answer is no. The bill of materials shows no third-party package runs in production, status checks the service live, and the continuity plan sets out what your customers keep if we stop and how a backup of the ledger is restored and checked: every day, automatically. Get a sandbox key Talk to us ## Usage in numbers · RiskRouter URL: https://riskrouter.eu/usage Usage counted live, from the ledger How much is recorded, and by whom Every figure on this page is counted from the quote ledger and the evidence log at the moment you load it, by your browser, from a public endpoint. Nothing here is typed in by us, so it cannot show traction we do not have. Where a number is zero, it says zero. Right now Entries, and who recorded them entries in both logs in the last 28 days recorded by keys that are not ours distinct keys recording, 28 days Count again The third figure is the one that matters. Everything else on this site can be produced by our own effort. An entry recorded with a key that is not ours means somebody outside decided the API was worth writing code against, ran it, and sent a key. Self-serve sandbox keys count as outside and are also shown separately below, because a key nobody had to ask us for is weaker evidence than a contracted one. Detail By log, and by origin of the key Figure Count Note Quote ledger (v1) entries Evidence log (v2) entries Entries by keys that are not ours of which self-serve sandbox keys Distributors with a key Twelve weeks Entries per calendar week Week starting (Monday) Quotes Evidence Not ours counting Adoption by people who are not us Witnesses and conformant implementations none yet independent witnesses none yet conformance results published by others Counted from witnesses/ and conformance-results/ in the repository when this page was built. A witness is listed only when it publishes its own key; a result only when its implementer publishes it themselves. Become a witness Run the conformance suite How this is counted What the figures mean, and what they do not Where the numbers come from GET /api/v1/usage , a public endpoint that counts the two logs on request and returns nothing but numbers and week dates: no identifier, no premium, no record. It is cached for five minutes at the edge. API reference . Ours Entries recorded with RiskRouter’s own keys, and entries that predate attribution. Our own keys are the validation key and those of the two test agents that use the live service every week as a prospect and as a customer would: a production key, and the sandbox key the prospect agent takes each week, tagged synthetic- . Their entries are in the totals above and never in the figure that matters. Everything else is not ours , including self-serve sandbox keys. What an entry is not Not a customer, not a policy, not revenue. It is one record sealed into a log. No cover is in force anywhere behind these figures; this is a validation build. Why publish it at all Because a number an outsider can fetch is worth more than a number on a slide, and because a small number said plainly is the only kind that can be believed when it grows. Every other claim we make, and where to check it . ## Every claim, and where to check it URL: https://riskrouter.eu/evidence Evidence for a security review, a regulator, or an investor Every claim, and where to check it This page exists so that due diligence takes an hour rather than a week. Each row is a claim we make somewhere on this site, the artefact that supports it, and who other than us can check it. Where the honest answer is nobody yet or no , the row says so. Nothing linked here needs an account, a call, or our cooperation. The record What the ledger and the evidence log prove Claim Check it at Who else can check Every recorded quote is chained; an altered, deleted or inserted entry is detectable /verify : the live head, and an in-browser checker for an export. Canonical form Anyone holding an export, with the public verifier , offline The head is signed, so we cannot later deny having published it Live attestation ; public key at /api/v1/ledger/pubkey and in the verifier repository Anyone, with verify-attestation.mjs or the Python verifier A signed head existed before a given date How old the record is : the anchors and their Bitcoin proofs, figures computed from the files at build time; proof-of-age.json Anyone with ots verify against the Bitcoin block The evidence log holds digests only, in an RFC 6962 Merkle tree, and never a record The specification , with test vectors ; the database refuses anything that is not a 64-hex digest Any implementer, with the vectors; any holder of a proof, with either verifier A chained record that was sealed and later held back shows as a gap; the log refuses a chain position out of turn Completeness chains ; the signed list at GET /api/v2/evidence/chain ; the construction Anyone given the chain’s tag, with chains.mjs or the Python verifier, offline. It cannot show that a record was chained at all One field of a sealed record can be shown and checked while the others stay sealed Selective disclosure ; the construction and its vectors Anyone holding the disclosure, with either verifier A spot-check sample cannot be steered by the firm or the auditor Spot checks ; the draw , with a vector seeded from Bitcoin’s first block Anyone, by repeating the draw; the block hash is checked on any Bitcoin node The formats are open and can be implemented and checked without us The conformance suite , in the kit at /kit/conformance/ Any implementer; their result is theirs to publish. None published by others yet Independent parties co-sign the log /witnesses ; keys in anchors/checkpoint/witnesses.json , cosignatures in anchors/checkpoint/cosigned/ Yes, since 23 October 2026: six staging witnesses of the public witness network cosign each new C2SP checkpoint. Nobody runs our own portable witness ( witnesses/ ) yet How much is recorded, and by whom /usage , counted live from the ledger by your browser Anyone; GET /api/v1/usage is public The service What runs, and how it is kept running Claim Check it at Who else can check The API is up and the ledger is reachable, right now /status , checked live by your browser, with the history measured from outside by GitHub Actions on a schedule of once every two hours that GitHub often runs late or skips, in practice every few hours; /uptime.json Anyone, on every page load What is deployed is what was reviewed build.commit on /api/v1/health names the commit CI deployed Anyone; the commit is on the deploying workflow’s run No third-party code runs in the API The bill of materials , derived from the Workers’ own imports on every build Anyone reading the SBOM The API is documented as it runs OpenAPI 3.1 , held identical to the routes by a test; reference Anyone generating a client from it Backups exist, restore, and match the anchored head Continuity : what each drill checks; daily and automatic Our records; a customer can ask for the latest drill report A customer can leave with their evidence intact The exit plan ; the export format and the verifiers are public Any customer, by exporting today Security and data What we hold, and what we do not Claim Check it at Who else can check A quote holds no personal data; an evidence entry is a salted digest of a record we never see Data protection , what the client sends Any integrator, by reading what their code sends Where data lives, and every sub-processor Sub-processors and Trust : EU hosting, plus one encrypted daily backup copy kept 90 days on GitHub, outside the EU The list is complete; there is no second list How to report a vulnerability, and the headers every page ships with security.txt (RFC 9116), Security Anyone, with a scanner The answers a vendor-risk team asks for Security questionnaire, answered , each answer linked to its evidence Any reviewer Whether the browser ever computes a price It never does: pricing is server-side only; the rating matrix and its fingerprint are published per version Anyone, by comparing the API’s answer with the published matrix What we do not claim The plain answers Registration RiskRouter is not registered with the FSMA. In our own analysis it is not an insurance intermediary: it is software for licensed distributors. A written question about the boundary was sent to the FSMA on 19 September 2026, and the FSMA replied by email on 29 September 2026, and again on 5 October 2026 to the same effect, pointing to its contact point for innovative business models. It restated the definitions of insurance intermediary and insurance distribution in the Act of 4 April 2014, and said that whether an activity needs registration, and under which status, is for the business to analyse itself, with a specialised legal adviser if need be, because the FSMA does not know its business model. It did not rule on ours, and its emails state that they are not official information from the FSMA. The boundary is therefore still our own analysis, not the FSMA’s; this page will say when a legal opinion settles it. Regulatory position . Customers, carriers, certifications None. No customer, carrier, certification or testimonial appears anywhere on this site, because none exists yet. A vendor’s name appears only once they have agreed to it in writing. Cover No cover is in force anywhere behind this system. Every price is a simulation of a published rating matrix, and every page says so. Uptime Measured from outside since 30 September 2026: GitHub Actions, which runs on none of our infrastructure, probes the site and the API on a schedule of once every two hours that GitHub often runs late or skips, in practice every few hours. Time it did not probe is shown as not measured, never as up. Status . What would change this page A witness key, a conformance result published by someone else, a first outside entry on /usage , or a specialised legal opinion on the registration question. Each is a fact, not a promise, and each changes a row above from none yet to a number or a link. ## Regime packs: what to record, rule by rule · RiskRouter URL: https://riskrouter.eu/regime-packs Regime packs one proof, mapped to each rule What to record, rule by rule The evidence log proves that a record existed, unaltered, no later than a signed and anchored head. It does not know what the record is. A pack says, for one regime, which fields to put in a record so that the proof answers the question a supervisor will ask, and maps each field to the article it speaks to. The record stays with you; only its salted fingerprint is sent. Fourteen packs, for fourteen laws, are published, as data anyone can use without us. How it works A record, its fingerprint, and the bundle you hand over 1. Write the record A JSON object with the pack's fields, in your own system. Identify people and cases by your own references, never by name: the pack's reference fields refuse anything that looks like a name or an email. 2. Send its fingerprint Canonicalise it (RFC 8785, with the profile in the specification ), hash it with a random salt, and send only the digest and the kind to POST /api/v2/evidence . Optionally sign it with your own key, so the leaf carries your signature too. 3. Keep three things The record, the salt, and the proof the API gives for the entry. Together they are a bundle : riskrouter-record-bundle|1 . 4. Hand over the bundle Anyone can check it offline with either verifier: the record hashes to the recorded digest, its kind is the recorded kind, the entry is in a tree whose head we signed, and, for a signed leaf, your own signature holds. # validate a record against its pack, then fingerprint it node records.mjs validate statement.json node records.mjs digest statement.json # prints salt_hex and record_digest; keep both # after recording: bundle it with the proof, and check it the way a supervisor would node records.mjs bundle statement.json --salt $SALT --proof proof.json > bundle.json node records.mjs check bundle.json --key signing-key.json python3 riskrouter_verify.py bundle.json --key signing-key.json records.mjs ships with the conformance suite , beside the packs, and in the public verifier repository ; the Python verifier is in the kit . The Python client's record_structured() does steps 2 and 3 for you. What the log proves What a bundle shows, and what it cannot It shows That this exact record, with these field values, existed no later than the head's time and the Bitcoin block its anchor is in, and has not been changed since. With a signed leaf, that you asserted it. A correction is a new record naming the old one in supersedes , and the old one stays: what was there before a correction can always be seen. It does not show That the record is true or complete, that an obligation applied to you, or that you met it. Whether a regime applies, what it requires of you, and how long to keep records are for you and your supervisor. Where the log adds most Where a rule asks that records cannot be altered or that logs are kept: Delegated Regulation (EU) 2017/565, Article 72(1), and the AI Act's Articles 12, 19 and 26(6). A database row can be edited by whoever administers it; a fingerprint in an anchored tree cannot be changed without the change showing. EU AI Act Anti-Money Laundering Regulation Consumer Credit Directive EU Clinical Trials Regulation Cyber Resilience Act DORA GDPR consent and erasure GDPR Article 22 Insurance Distribution Directive EU Medical Device Regulations (MDR and IVDR) MiFID II NIS2 and data breaches Public procurement VAT Directive invoices EU AI Act pack ai-act version 1 High-risk AI systems: logs, decisions and human oversight What a provider or deployer of a high-risk AI system records so that its logs, the decisions taken on the system's output, and the human oversight exercised over it can be shown later to be exactly what existed at the time; and what a provider of a high-risk system or of a general-purpose model records of the data sets a model was trained, validated and tested on. The law Regulation (EU) 2024/1689 (Artificial Intelligence Act) (CELEX 32024R1689) Kinds ai.log-segment log segment sealed ai.decision decision taken on the system's output ai.oversight human oversight exercised ai.model-provenance model, prompt and policy in force ai.screening-criteria criteria fixed before people are assessed ai.content-credential generated content and its marking sealed ai.training-data training, validation or test data set fixed The file /packs/ai-act.json , with an example record for every kind How long to keep it Articles 19(1) and 26(6) set a minimum of six months for the logs they cover, unless other law provides otherwise; this pack does not decide your period. Whether your system is high-risk, and from when these obligations apply to it, is yours to establish from the Regulation. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2024/1689, Article 12(1) High-risk AI systems must technically allow for the automatic recording of events (logs) over the lifetime of the system. ai.log-segment : system_ref, segment_start, segment_end, event_count, log_digest, previous_segment_record_id ai.model-provenance : system_ref, effective_from Regulation (EU) 2024/1689, Article 12(2) Logging must enable recording of events relevant for identifying situations that may result in the system presenting a risk or in a substantial modification, for post-market monitoring (Article 72), and for monitoring the system's operation (Article 26(5)). ai.log-segment : system_version ai.decision : system_ref, system_version, input_digest, output_digest, decided_at ai.model-provenance : system_ref, model_version_digest, prompt_template_digest, policy_digest, effective_from, previous_record_id ai.screening-criteria : criteria_digest, provenance_record_id Regulation (EU) 2024/1689, Article 14(4)(d) The people overseeing the system must be able to decide, in any particular situation, not to use it or to disregard, override or reverse its output. ai.decision : human_involved ai.oversight : system_ref, decision_record_id, overseer_ref, action, reason, acted_at Regulation (EU) 2024/1689, Article 19(1) Providers keep the logs their high-risk AI systems generate automatically, to the extent the logs are under their control, for a period appropriate to the system's intended purpose and of at least six months, unless Union or national law, in particular on personal data, provides otherwise. ai.log-segment : role, log_digest, retain_until Regulation (EU) 2024/1689, Article 19(2) Providers that are financial institutions subject to internal-governance requirements under Union financial services law keep those logs as part of the documentation kept under that law. ai.log-segment : retain_until Regulation (EU) 2024/1689, Article 26(6) Deployers keep the logs the system generates automatically, to the extent the logs are under their control, for a period appropriate to the system's intended purpose and of at least six months, unless applicable law provides otherwise. ai.log-segment : role, log_digest, retain_until Regulation (EU) 2024/1689, Article 86(1) A person affected by a decision a deployer takes on the basis of output from a high-risk AI system listed in Annex III (other than point 2), with legal or similarly significant adverse effects, may obtain clear and meaningful explanations of the role of the AI system in the decision and of the main elements of the decision taken. ai.decision : system_ref, case_ref, output_summary, role_of_system, decision ai.model-provenance : model_version_digest, prompt_template_digest, policy_digest ai.screening-criteria : system_ref, round_ref, criteria_digest, fixed_at, opens_at Regulation (EU) 2024/1689, Article 50(2) Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, as far as technically feasible. ai.content-credential : system_ref, content_digest, marking, marking_digest, generated_at Regulation (EU) 2024/1689, Article 10(2) Training, validation and testing data sets of high-risk AI systems are subject to data governance and management practices appropriate to the intended purpose, concerning in particular the data collection processes and the origin of the data (and, for personal data, the original purpose of the collection) and the data-preparation operations such as annotation, labelling, cleaning, updating, enrichment and aggregation. ai.training-data : dataset_ref, role, use, manifest_digest, item_count, origin_record_digest, preparation_record_digest, fixed_at, previous_record_id Regulation (EU) 2024/1689, Article 53(1)(a) Providers of general-purpose AI models draw up and keep up to date the technical documentation of the model, including its training and testing process and the results of its evaluation, to provide it on request to the AI Office and the national competent authorities. ai.training-data : dataset_ref, role, manifest_digest, fixed_at Regulation (EU) 2024/1689, Article 53(1)(c) Providers of general-purpose AI models put in place a policy to comply with Union law on copyright and related rights, in particular to identify and comply with a reservation of rights expressed under Article 4(3) of Directive (EU) 2019/790. ai.training-data : copyright_check_digest Regulation (EU) 2024/1689, Article 53(1)(d) Providers of general-purpose AI models draw up and make publicly available a sufficiently detailed summary about the content used for training the model, according to a template provided by the AI Office. ai.training-data : public_summary_digest ai.log-segment Log segment sealed When: Each time a segment of the system's automatic log is closed: every hour, every day, or every file rotation. Field What to put in it Type Mapped to system_ref Your reference for the AI system. your reference Article 12(1) system_version The version of the system that produced the log. one line Article 12(2) role Whether you keep this log as the provider or as a deployer. one of provider , deployer Article 19(1) Article 26(6) segment_start Time of the first event in the segment, UTC. UTC time Article 12(1) segment_end Time of the last event in the segment, UTC. UTC time Article 12(1) event_count How many events the segment holds, so a shortened log is caught. whole number Article 12(1) log_digest SHA-256 of the segment file exactly as stored. SHA-256 Article 12(1) Article 19(1) Article 26(6) log_format (optional) The format of the segment, such as jsonl. one line previous_segment_record_id (optional) The record_id of the segment before this one, so a missing segment is caught. your reference Article 12(1) retain_until (optional) The date until which you have decided to keep this segment. date Article 19(1) Article 26(6) Article 19(2) ai.decision Decision taken on the system's output When: When a decision about a person or a case is taken on the basis of the system's output. Field What to put in it Type Mapped to system_ref Your reference for the AI system. your reference Article 12(2) Article 86(1) system_version The version of the system that produced the output. one line Article 12(2) case_ref Your reference for the case or the person concerned. Never a name. your reference Article 86(1) input_digest (optional) SHA-256 of the input exactly as the system received it. SHA-256 Article 12(2) output_digest (optional) SHA-256 of the output exactly as the system produced it. SHA-256 Article 12(2) output_summary What the system output, in words someone can check against the output. text Article 86(1) role_of_system The role the system's output played in the decision. text Article 86(1) decision The decision taken, and its main elements. text Article 86(1) human_involved Whether a person reviewed the output before the decision. true or false Article 14(4)(d) decided_at When the decision was taken, UTC. UTC time Article 12(2) ai.oversight Human oversight exercised When: When a person overseeing the system confirms, overrides, reverses or declines to use its output, or stops it. Field What to put in it Type Mapped to system_ref Your reference for the AI system. your reference Article 14(4)(d) decision_record_id (optional) The record_id of the ai.decision record concerned, if any. your reference Article 14(4)(d) overseer_ref Your reference for the person who acted. Never a name. your reference Article 14(4)(d) action What the person did with the output. one of confirmed , overridden , reversed , not-used , stopped Article 14(4)(d) reason Why. text Article 14(4)(d) acted_at When, UTC. UTC time Article 14(4)(d) ai.model-provenance Model, prompt and policy in force When: Each time the model version, the prompt template or the policy a system applies changes: sealed by their digests, so any later decision record can name exactly what was in force. Field What to put in it Type Mapped to system_ref Your reference for the AI system. your reference Article 12(1) Article 12(2) model_version_digest SHA-256 of the model identifier and version, or of the model weights, as you define it. SHA-256 Article 12(2) Article 86(1) prompt_template_digest (optional) SHA-256 of the prompt template or system instructions, exactly as deployed. SHA-256 Article 12(2) Article 86(1) policy_digest (optional) SHA-256 of the rules or policy the system applies, exactly as deployed. SHA-256 Article 12(2) Article 86(1) effective_from From when this combination is in force, UTC. UTC time Article 12(1) Article 12(2) previous_record_id (optional) The record_id of the provenance record this one replaces. your reference Article 12(2) ai.screening-criteria Criteria fixed before people are assessed When: Before a round of assessment opens (applications for a job, for credit, for a place): the criteria and their weights are sealed, so they can be shown not to have been changed once the people being assessed were known. Field What to put in it Type Mapped to system_ref Your reference for the AI system or the assessment process. your reference Article 86(1) round_ref Your reference for the round, such as a vacancy. your reference Article 86(1) criteria_digest SHA-256 of the criteria and their weights, exactly as fixed. SHA-256 Article 86(1) Article 12(2) fixed_at When they were fixed, UTC. UTC time Article 86(1) opens_at (optional) When the round opens, UTC. UTC time Article 86(1) provenance_record_id (optional) The record_id of the ai.model-provenance in force. your reference Article 12(2) ai.content-credential Generated content and its marking sealed When: When a system generates content and marks it: the digest of the content and of its marking (for example a C2PA manifest) are sealed, so the time it existed is fixed independently and two different versions of it cannot both be presented as the original. Field What to put in it Type Mapped to system_ref Your reference for the generating system. your reference Article 50(2) content_digest SHA-256 of the generated content exactly as published. SHA-256 Article 50(2) marking How the output is marked. one of c2pa-manifest , watermark , metadata , other Article 50(2) marking_digest (optional) SHA-256 of the marking itself, such as the C2PA manifest store, where it can be separated. SHA-256 Article 50(2) generated_at When it was generated, UTC. UTC time Article 50(2) ai.training-data Training, validation or test data set fixed When: Each time a data set used to train, validate or test a model is fixed: sealed by the fingerprint of its manifest (every file and its SHA-256), with the records of where the data came from and how it was prepared, so a later question about which data a model learned from has one answer. Field What to put in it Type Mapped to dataset_ref Your reference for the data set and its version. your reference Article 10(2) Article 53(1)(a) role Whether you keep this as the provider of a high-risk AI system or of a general-purpose AI model. one of high-risk-provider , gpai-provider Article 10(2) Article 53(1)(a) use What the data set is used for. one of training , validation , testing Article 10(2) manifest_digest SHA-256 of the manifest listing every file of the data set with its own SHA-256, exactly as stored. SHA-256 Article 10(2) Article 53(1)(a) item_count How many items the data set holds, so a later, smaller or larger set is caught. whole number Article 10(2) origin_record_digest SHA-256 of your record of the collection processes and the origin of the data. SHA-256 Article 10(2) preparation_record_digest (optional) SHA-256 of your record of the preparation operations: annotation, labelling, cleaning, updating, enrichment, aggregation. SHA-256 Article 10(2) copyright_check_digest (optional) SHA-256 of your record of how reservations of rights were identified and honoured for this data set. SHA-256 Article 53(1)(c) public_summary_digest (optional) SHA-256 of the published summary of training content, exactly as published. SHA-256 Article 53(1)(d) fixed_at When the data set was fixed in this form, UTC. UTC time Article 10(2) Article 53(1)(a) previous_record_id (optional) The record_id of the training-data record for the version this one replaces. your reference Article 10(2) The log itself never leaves you: only the digest of each segment is recorded. A segment produced later can then be shown to be the segment that existed when it was sealed, and event_count and previous_segment_record_id make a shortened or missing segment visible. Anti-Money Laundering Regulation pack aml version 1 Anti-money laundering: customer due diligence, sanctions screening and its records What an obliged entity records so it can show which customer file it approved and when, which version of a targeted financial sanctions list a customer was screened against, and that the screening was repeated. The customer file stays with you; only its digests are sealed. The law Regulation (EU) 2024/1624 (Anti-Money Laundering Regulation) (CELEX 32024R1624) Kinds aml.onboarding-file customer file approved aml.list-version sanctions list version held aml.screening customer screened against a sanctions list The file /packs/aml.json , with an example record for every kind How long to keep it Article 77 states five years from the end of the business relationship, or from the date of an occasional transaction, and says when that period may be extended. The regulation applies from 10 July 2027. This pack does not decide your period. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2024/1624, Article 20(1) The customer due diligence measures: among them, identifying the customer and the beneficial owners and verifying their identity, and understanding the purpose and intended nature of the business relationship. aml.onboarding-file : customer_ref, approved_at, cdd_level, document_digests, approver_ref Regulation (EU) 2024/1624, Article 20(1), point (d) Verify whether the customer or the beneficial owners are subject to targeted financial sanctions and, for a legal entity, whether persons subject to them control it or hold more than 50 % of its proprietary rights or a majority interest in it. aml.onboarding-file : screening_record_ids aml.list-version : list_ref, list_digest aml.screening : customer_ref, list_record_id, trigger, screened_at, result, reviewer_ref Regulation (EU) 2024/1624, Article 26 Regularly verify again whether the condition in Article 20(1), point (d), is met, as often as the exposure to the risks of non-implementation and evasion of targeted financial sanctions requires; credit and financial institutions also do so upon a new designation. aml.list-version : list_ref, published_at, retrieved_at, list_digest aml.screening : customer_ref, list_record_id, trigger, screened_at Regulation (EU) 2024/1624, Article 77(1) Retain, among other things, a copy of the documents and information obtained in performing customer due diligence, the record of the assessment of suspicious transactions, and the supporting evidence and records of transactions. aml.onboarding-file : customer_ref, document_digests aml.onboarding-file Customer file approved When: When a customer's due diligence file is approved: the digest of each document in it is sealed at that moment, so no document can be swapped in later. Field What to put in it Type Mapped to customer_ref Your reference for the customer, never a name. your reference Article 20(1) Article 77(1) approved_at When the file was approved, UTC. UTC time Article 20(1) cdd_level The level of due diligence applied. one of simplified , standard , enhanced Article 20(1) document_digests SHA-256 of each document and data set in the file, exactly as held. list of SHA-256 Article 77(1) Article 20(1) screening_record_ids (optional) The record_ids of the sanctions screenings done for this file. list of your reference Article 20(1), point (d) approver_ref Your reference for the person who approved the file. your reference Article 20(1) aml.list-version Sanctions list version held When: Each time you take in a new version of a targeted financial sanctions list: its digest and publication date are sealed, so a screening can later be tied to the list as it stood. Field What to put in it Type Mapped to list_ref Your reference for the list, such as eu-consolidated or un-sc. your reference Article 20(1), point (d) Article 26 published_at When the list version was published by its source, UTC. UTC time Article 26 retrieved_at When you retrieved it, UTC. UTC time Article 26 list_digest SHA-256 of the list file exactly as retrieved. SHA-256 Article 20(1), point (d) Article 26 aml.screening Customer screened against a sanctions list When: At onboarding, at each periodic re-check, and upon a new designation. Field What to put in it Type Mapped to customer_ref Your reference for the customer, never a name. your reference Article 20(1), point (d) Article 26 list_record_id The record_id of the aml.list-version screened against. your reference Article 20(1), point (d) Article 26 trigger Why the screening was done. one of onboarding , periodic , new-designation , change-of-customer-data Article 20(1), point (d) Article 26 screened_at When, UTC. UTC time Article 20(1), point (d) Article 26 result The outcome. one of no-match , potential-match-cleared , potential-match-open , confirmed-match Article 20(1), point (d) reviewer_ref (optional) Your reference for the person who cleared or confirmed a potential match. your reference Article 20(1), point (d) The customer file, the list and the screening detail stay with you; only their digests are sealed. A sealed screening shows which list version was checked and when, not that the check was right. Consumer Credit Directive pack ccd version 1 Consumer credit: the creditworthiness assessment and the right to human intervention What a creditor records so it can show that a thorough creditworthiness assessment was made before the credit agreement, on which data and with which model, and that a consumer who asked for human intervention, an explanation or a review got one. The law Directive (EU) 2023/2225 (Consumer Credit Directive) (CELEX 32023L2225) Kinds credit.creditworthiness-assessment creditworthiness assessed credit.human-intervention human intervention, explanation or review given The file /packs/ccd.json , with an example record for every kind How long to keep it Article 18 does not state how long to keep these records, and this pack does not decide it for you. The directive applies from 20 November 2026 as transposed in each Member State; check your national law. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Directive (EU) 2023/2225, Article 18(1) Before concluding a credit agreement, the creditor carries out a thorough assessment of the consumer's creditworthiness, in the consumer's interest, to prevent irresponsible lending practices and over-indebtedness. credit.creditworthiness-assessment : application_ref, assessed_at, inputs_digest, model_version_digest, outcome, assessor_ref Directive (EU) 2023/2225, Article 18(8) Where the assessment involves automated processing of personal data, the consumer has the right to request and obtain human intervention on the part of the creditor, an explanation of the assessment, and to express their point of view and request a review of the assessment and of the decision. credit.creditworthiness-assessment : model_version_digest, automated credit.human-intervention : application_ref, assessment_record_id, requested, requested_at, answered_at, reviewer_ref, decision_changed, explanation_digest credit.creditworthiness-assessment Creditworthiness assessed When: When the assessment is made, before the agreement is concluded: the inputs and the model are sealed by their digests, so the assessment can be shown to have come first. Field What to put in it Type Mapped to application_ref Your reference for the credit application. your reference Article 18(1) assessed_at When the assessment was made, UTC. UTC time Article 18(1) inputs_digest SHA-256 of the data the assessment used, exactly as used (for example, a manifest of the documents and figures). SHA-256 Article 18(1) model_version_digest (optional) SHA-256 of the scoring model or rule set used, if one was. SHA-256 Article 18(1) Article 18(8) automated Whether the assessment involved automated processing of personal data. true or false Article 18(8) outcome The result of the assessment. one of positive , negative , referred Article 18(1) assessor_ref (optional) Your reference for the person who made or approved the assessment, if a person did. your reference Article 18(1) credit.human-intervention Human intervention, explanation or review given When: Each time a consumer asks for human intervention, an explanation or a review of an automated assessment, and when it is given. Field What to put in it Type Mapped to application_ref Your reference for the credit application. your reference Article 18(8) assessment_record_id The record_id of the assessment concerned. your reference Article 18(8) requested What the consumer asked for. list of human-intervention , explanation , review Article 18(8) requested_at When the consumer asked, UTC. UTC time Article 18(8) answered_at When it was given, UTC. UTC time Article 18(8) reviewer_ref Your reference for the person who intervened or reviewed. your reference Article 18(8) decision_changed Whether the review changed the decision. true or false Article 18(8) explanation_digest (optional) SHA-256 of the explanation given, exactly as given. SHA-256 Article 18(8) Mortgage credit is outside this directive. Where the assessment uses a high-risk AI system, the AI Act pack covers its logs; where a decision is based solely on automated processing, the GDPR Article 22 pack covers the safeguards that apply. EU Clinical Trials Regulation pack clinical-trials version 1 Clinical trials: the trial master file, safety reports and serious breaches What a sponsor or an investigator records so that the essential documents of a clinical trial, the safety reports exchanged during it and the archive kept after it can be shown later to be exactly what existed at the time, and any alteration to be traceable. The law Regulation (EU) No 536/2014 (Clinical Trials Regulation) (CELEX 32014R0536) Kinds clinical.tmf-document essential document filed in the trial master file clinical.tmf-archive trial master file archived clinical.adverse-event-report adverse event report sent or received clinical.serious-breach-notification serious breach notified The file /packs/clinical-trials.json , with an example record for every kind How long to keep it Article 58 sets at least 25 years after the end of the trial for the content of the clinical trial master file, unless other Union law requires longer; subjects' medical files follow national law. This pack does not decide your period. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) No 536/2014, Article 41(2) The investigator records and documents all adverse events, unless the protocol provides differently, and reports serious adverse events to the sponsor without undue delay and not later than 24 hours after becoming aware of them, unless the protocol provides that no immediate reporting is required, with a follow-up report where relevant. clinical.adverse-event-report : trial_ref, report_ref, role, seriousness, report_type, aware_at, reported_at, report_digest Regulation (EU) No 536/2014, Article 41(3) The sponsor keeps detailed records of all adverse events reported to it by the investigator. clinical.adverse-event-report : trial_ref, report_ref, role, reported_at, report_digest Regulation (EU) No 536/2014, Article 52(1) The sponsor notifies the Member States concerned, through the EU portal, of a serious breach of the Regulation or of the protocol version applicable at the time, without undue delay and not later than seven days after becoming aware of it. clinical.serious-breach-notification : trial_ref, breach_ref, aware_at, notified_at, notification_digest Regulation (EU) No 536/2014, Article 57 The sponsor and the investigator each keep a clinical trial master file that at all times contains the essential documents allowing the conduct of the trial and the quality of the data generated to be verified, readily available and directly accessible to the Member States on request. clinical.tmf-document : trial_ref, holder, site_ref, document_type, document_version, document_digest, filed_at Regulation (EU) No 536/2014, Article 58, first paragraph Unless other Union law requires longer, the sponsor and the investigator archive the content of the clinical trial master file for at least 25 years after the end of the trial; the medical files of subjects are archived in accordance with national law. clinical.tmf-archive : trial_ref, holder, trial_end_date, archived_at, retain_until Regulation (EU) No 536/2014, Article 58, fifth paragraph The media used to archive the content of the clinical trial master file keep it complete and legible throughout the archiving period. clinical.tmf-archive : manifest_digest, file_count, reason Regulation (EU) No 536/2014, Article 58, sixth paragraph Any alteration to the content of the clinical trial master file is traceable. clinical.tmf-document : document_version, document_digest, previous_version_record_id clinical.tmf-archive : manifest_digest, reason, previous_archive_record_id clinical.tmf-document Essential document filed in the trial master file When: Each time an essential document, or a new version of one, is filed: a protocol version, an informed consent form, a monitoring report, a signed agreement. Field What to put in it Type Mapped to trial_ref The trial's EU trial number, or your own reference for the trial. your reference Article 57 holder Whose trial master file this is: the sponsor's or an investigator's. one of sponsor , investigator Article 57 site_ref (optional) Your reference for the trial site, for an investigator's file. your reference Article 57 document_type What the document is, in your own taxonomy, such as protocol or monitoring-report. one line Article 57 document_version The document's version, as written on it. one line Article 57 Article 58, sixth paragraph document_digest SHA-256 of the document file exactly as filed. SHA-256 Article 57 Article 58, sixth paragraph filed_at When it was filed, UTC. UTC time Article 57 previous_version_record_id (optional) The record_id of the version this one replaces, so every alteration is traceable and the earlier version stays sealed. your reference Article 58, sixth paragraph clinical.tmf-archive Trial master file archived When: When the content of a trial master file is archived after the end of the trial, and again when it is moved to new media or transferred to a new owner. Field What to put in it Type Mapped to trial_ref The trial's EU trial number, or your own reference for the trial. your reference Article 58, first paragraph holder Whose trial master file this is: the sponsor's or an investigator's. one of sponsor , investigator Article 58, first paragraph trial_end_date The end of the trial, from which the archiving period runs. date Article 58, first paragraph archived_at When the archive was closed, UTC. UTC time Article 58, first paragraph manifest_digest SHA-256 of the archive's file list, one line per file with its own SHA-256, such as the list a folder seal produces. SHA-256 Article 58, fifth paragraph Article 58, sixth paragraph file_count How many files the archive holds, so a file left out shows. whole number Article 58, fifth paragraph reason Why this archive record was made. one of archived , media-migration , ownership-transfer Article 58, fifth paragraph Article 58, sixth paragraph previous_archive_record_id (optional) The record_id of the archive record this one follows, after a migration or a transfer. your reference Article 58, sixth paragraph retain_until (optional) The date until which you have decided to keep the archive. date Article 58, first paragraph clinical.adverse-event-report Adverse event report sent or received When: Each time an investigator sends, or a sponsor receives, a report of an adverse event: the initial report and each follow-up. Field What to put in it Type Mapped to trial_ref The trial's EU trial number, or your own reference for the trial. your reference Article 41(2) Article 41(3) report_ref Your reference for the report; never the subject's name or a direct identifier. your reference Article 41(2) Article 41(3) role Whether you sent the report as investigator or received it as sponsor. one of investigator-sent , sponsor-received Article 41(2) Article 41(3) seriousness Whether the event was reported as serious. one of serious , non-serious Article 41(2) report_type Whether this is the first report or a follow-up. one of initial , follow-up Article 41(2) aware_at (optional) When the investigator became aware of the event, UTC; the 24-hour period runs from here. UTC time Article 41(2) reported_at When the report was sent or received, UTC. UTC time Article 41(2) Article 41(3) report_digest SHA-256 of the report exactly as sent or received. SHA-256 Article 41(2) Article 41(3) clinical.serious-breach-notification Serious breach notified When: When the sponsor notifies a serious breach through the EU portal. Field What to put in it Type Mapped to trial_ref The trial's EU trial number, or your own reference for the trial. your reference Article 52(1) breach_ref Your reference for the breach. your reference Article 52(1) aware_at When the sponsor became aware of the breach, UTC; the seven days run from here. UTC time Article 52(1) notified_at When the notification was submitted, UTC. UTC time Article 52(1) notification_digest SHA-256 of the notification exactly as submitted. SHA-256 Article 52(1) Only digests are sent; the documents, the reports and anything about a subject stay with you. A new version is a new record that names the one it replaces, and the earlier one stays sealed, so an alteration shows as a chain of records rather than an overwrite. A record evidences what existed and when; whether a document is an essential document, and whether a deadline was met, is for the sponsor and the investigator to establish. Cyber Resilience Act pack cra version 1 Cyber Resilience Act: what shipped, its components, and when a vulnerability was reported What a manufacturer of a product with digital elements records so it can show what it released, with which components, and that each report about an actively exploited vulnerability was sent within its deadline. The release and the reports stay with you; only their digests are sealed. The law Regulation (EU) 2024/2847 (Cyber Resilience Act) (CELEX 32024R2847) Kinds cra.release release made available cra.vulnerability-report report on an actively exploited vulnerability sent The file /packs/cra.json , with an example record for every kind How long to keep it The regulation requires the technical documentation to be kept for at least ten years after the product is placed on the market, or for the support period if that is longer. Article 14 has applied since 11 September 2026; most other obligations apply from 11 December 2027. This pack does not decide your period. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2024/2847, Article 13 Manufacturers place products with digital elements on the market only if they meet the essential cybersecurity requirements of Annex I, whose Part II includes identifying and documenting the product's vulnerabilities and components, among other things by drawing up a software bill of materials. cra.release : product_ref, sbom_digest, sbom_format, support_period_end Regulation (EU) 2024/2847, Article 31 The technical documentation, with the contents Annex VII sets out, is drawn up before the product is placed on the market and kept up to date. cra.release : product_ref, version, released_at, artefact_digest, sbom_digest Regulation (EU) 2024/2847, Article 14(1) Notify any actively exploited vulnerability in the product that the manufacturer becomes aware of, simultaneously to the CSIRT designated as coordinator and to ENISA, via the single reporting platform. cra.vulnerability-report : product_ref, vulnerability_ref, aware_at, submitted_at, report_digest Regulation (EU) 2024/2847, Article 14(2) An early warning within 24 hours of becoming aware, a vulnerability notification within 72 hours, and a final report no later than 14 days after a corrective or mitigating measure is available. cra.vulnerability-report : vulnerability_ref, stage, aware_at, corrective_measure_at, submitted_at, report_digest cra.release Release made available When: Each time a version is released: the artefact and its software bill of materials are sealed by their digests at that moment. A SCITT Signed Statement (POST /api/v2/evidence/statement) does the same for a supply chain that already speaks SCITT. Field What to put in it Type Mapped to product_ref Your reference for the product. your reference Article 13 Article 31 version The version released. one line Article 31 released_at When it was made available, UTC. UTC time Article 31 artefact_digest SHA-256 of the released artefact exactly as shipped. SHA-256 Article 31 sbom_digest SHA-256 of the software bill of materials for this release. SHA-256 Article 13 Article 31 sbom_format (optional) The format of the bill of materials. one of cyclonedx , spdx , other Article 13 support_period_end (optional) The end of the support period you declared for this product. date Article 13 cra.vulnerability-report Report on an actively exploited vulnerability sent When: Each time something is sent: the early warning, the vulnerability notification, the final report. The deadlines run from aware_at, and for the final report from the corrective measure. Field What to put in it Type Mapped to product_ref Your reference for the product. your reference Article 14(1) vulnerability_ref Your reference for the vulnerability, or its CVE identifier. your reference Article 14(1) Article 14(2) stage What was sent. one of early-warning , vulnerability-notification , final-report Article 14(2) aware_at When you became aware of the active exploitation, UTC. UTC time Article 14(1) Article 14(2) corrective_measure_at (optional) When a corrective or mitigating measure became available, UTC; the final report's deadline runs from here. UTC time Article 14(2) submitted_at When it was sent, UTC. UTC time Article 14(1) Article 14(2) report_digest SHA-256 of what was sent, exactly as sent. SHA-256 Article 14(1) Article 14(2) Whether a product is in scope, and in which class, is for the manufacturer to establish from the regulation. `node tools/deadline.mjs` checks the times of sealed reports against the 24-hour, 72-hour and 14-day deadlines. DORA pack dora version 1 ICT incidents, incident reports and the register of information What a financial entity records so that its incident log, the reports it sent to its competent authority and its register of ICT third-party arrangements can be shown later to be exactly what existed at each date. The law Regulation (EU) 2022/2554 (Digital Operational Resilience Act) (CELEX 32022R2554) Kinds dora.incident incident recorded at a stage dora.incident-report report submitted to the competent authority dora.register-snapshot register of information at a date The file /packs/dora.json , with an example record for every kind How long to keep it This pack does not state how long to keep these records. Check the period with your competent authority. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2022/2554, Article 17(1) Define, establish and implement an ICT-related incident management process to detect, manage and notify ICT-related incidents. dora.incident : stage, detected_at, stage_at Regulation (EU) 2022/2554, Article 17(2) Record all ICT-related incidents and significant cyber threats, with procedures for consistent monitoring, handling and follow-up so root causes are identified, documented and addressed. dora.incident : incident_ref, category, description, root_cause, actions Regulation (EU) 2022/2554, Article 18 Classify ICT-related incidents and determine their impact against the criteria the article sets. dora.incident : classification, services_affected Regulation (EU) 2022/2554, Article 19(4) For a major ICT-related incident, submit to the competent authority an initial notification, an intermediate report and a final report. dora.incident-report : incident_ref, report_type, authority, report_digest, submission_ref, submitted_at Regulation (EU) 2022/2554, Article 28(3) Maintain and update a register of information on all contractual arrangements for ICT services provided by ICT third-party service providers. dora.register-snapshot : as_of, register_digest, arrangements, purpose dora.incident Incident recorded at a stage When: At each stage of an ICT-related incident or significant cyber threat: detected, classified, resolved, closed. Field What to put in it Type Mapped to incident_ref Your reference for the incident. your reference Article 17(2) stage The stage this record captures. one of detected , classified , resolved , closed Article 17(1) category Whether it is an incident or a significant cyber threat. one of ict-related-incident , significant-cyber-threat Article 17(2) classification Your classification under Article 18 at this stage. one of major , not-major , not-yet-classified Article 18 services_affected (optional) Your references for the services affected. list of your reference Article 18 description What happened, as known at this stage. text Article 17(2) root_cause (optional) The root cause, once identified. text Article 17(2) actions (optional) What was done, and what will be done to prevent recurrence. text Article 17(2) detected_at When the incident was detected, UTC. UTC time Article 17(1) stage_at When this stage was reached, UTC. UTC time Article 17(1) dora.incident-report Report submitted to the competent authority When: When an initial notification, an intermediate report or a final report is submitted. Field What to put in it Type Mapped to incident_ref Your reference for the incident. your reference Article 19(4) report_type Which report. one of initial-notification , intermediate-report , final-report Article 19(4) authority The competent authority it went to. one line Article 19(4) report_digest SHA-256 of the report exactly as submitted. SHA-256 Article 19(4) submission_ref (optional) The authority's acknowledgement reference, if any. your reference Article 19(4) submitted_at When it was submitted, UTC. UTC time Article 19(4) dora.register-snapshot Register of information at a date When: Each time the register is updated, and when it is submitted. Field What to put in it Type Mapped to as_of The date the register describes. date Article 28(3) register_digest SHA-256 of the register file exactly as kept or submitted. SHA-256 Article 28(3) arrangements How many contractual arrangements it lists. whole number Article 28(3) file_format (optional) The format of the file, as your authority names it. one line purpose Whether this is an internal update or the version submitted. one of update , submission Article 28(3) The reports themselves go to your authority by its own channel; the log never sends anything to anyone. What the log adds is that the report you hold today can be shown to be the one you held on the day you submitted it. GDPR consent and erasure pack gdpr-accountability version 1 GDPR accountability: consent given and withdrawn, and erasure carried out What a controller records so it can show that a person consented, to which notice and for which purpose, that a withdrawal was honoured, and that a record it had to erase was erased on time. The records hold no personal data beyond your own references, and the personal data they refer to stays erasable in your systems. The law Regulation (EU) 2016/679 (General Data Protection Regulation) (CELEX 32016R0679) Kinds gdpr.consent consent given or withdrawn gdpr.erasure erasure carried out The file /packs/gdpr-accountability.json , with an example record for every kind How long to keep it The GDPR states no fixed period for these records: keep them as long as you may need to demonstrate the consent or the erasure, and no longer. This pack does not decide it for you. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2016/679, Article 5(2) The controller is responsible for, and must be able to demonstrate, compliance with the principles in Article 5(1). gdpr.consent : subject_ref, notice_digest gdpr.erasure : subject_ref, erased_at, erased_record_digests, systems Regulation (EU) 2016/679, Article 7(1) Where processing is based on consent, the controller must be able to demonstrate that the person consented to the processing of their personal data. gdpr.consent : subject_ref, event, purposes, notice_digest, notice_version, channel, at Regulation (EU) 2016/679, Article 7(3) The person may withdraw consent at any time, withdrawing must be as easy as giving consent, and the person is informed of this before giving it. gdpr.consent : event, channel, at Regulation (EU) 2016/679, Article 17(1) The person has the right to obtain the erasure of personal data concerning them without undue delay, and the controller must erase it without undue delay, where one of the grounds the article lists applies. gdpr.erasure : subject_ref, ground, requested_at, erased_at, systems gdpr.consent Consent given or withdrawn When: Each time a person gives consent, and each time they withdraw it: one record per event, the notice they saw sealed by its digest. Field What to put in it Type Mapped to subject_ref Your own reference for the person, never their name or email. your reference Article 5(2) Article 7(1) event Whether consent was given or withdrawn. one of given , withdrawn Article 7(1) Article 7(3) purposes The purposes consented to, as your own short codes. list of one line Article 7(1) notice_digest SHA-256 of the notice exactly as the person saw it, so the wording they agreed to can be shown later. SHA-256 Article 7(1) Article 5(2) notice_version (optional) Your version label for that notice. one line Article 7(1) channel Where the consent was given or withdrawn. one of web , app , email , paper , phone , in-person , other Article 7(1) Article 7(3) at When, UTC. UTC time Article 7(1) Article 7(3) gdpr.erasure Erasure carried out When: When personal data is erased on request or because its period has run out. The record that was erased can be referred to by its record_digest: the digest stays, the data goes. Field What to put in it Type Mapped to subject_ref Your own reference for the person, never their name or email. your reference Article 5(2) Article 17(1) ground Why the data was erased. one of request , consent-withdrawn , objection , no-longer-necessary , unlawful , legal-obligation , retention-expired Article 17(1) requested_at (optional) When the person asked, UTC, if they did. UTC time Article 17(1) erased_at When the erasure was completed, UTC. UTC time Article 17(1) Article 5(2) erased_record_digests (optional) The record_digests of sealed records whose underlying data was erased. Each seal stays and now proves only that a record existed. list of SHA-256 Article 5(2) systems Your references for the systems the data was erased from. list of your reference Article 5(2) Article 17(1) The records name people only by your own references; the personal data stays in your systems and erasable there. Recording a consent does not make it valid: whether consent was freely given, specific, informed and unambiguous is decided by the facts, not by the seal. GDPR Article 22 pack gdpr-art22 version 1 Automated individual decisions: basis, information and human intervention What a controller records to show, for a decision based solely on automated processing, which exception it relied on, what the person was told, and what happened when they asked for a human to look again. The law Regulation (EU) 2016/679 (General Data Protection Regulation) (CELEX 32016R0679) Kinds gdpr.automated-decision automated decision taken gdpr.human-intervention human intervention requested and handled The file /packs/gdpr-art22.json , with an example record for every kind How long to keep it This pack does not state how long to keep these records. The records are yours and hold personal data under your control: keep them no longer than your purposes need, and they stay erasable because they never leave you. Only a salted digest is in the log, and without the record and its salt nobody, including us, can link it to anyone. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2016/679, Article 5(2) The controller is responsible for, and must be able to demonstrate, compliance with the principles in Article 5(1). gdpr.automated-decision : basis_ref, outcome, decided_at gdpr.human-intervention : reasons Regulation (EU) 2016/679, Article 13(2)(f) and Article 14(2)(g) Inform the person of the existence of automated decision-making, including profiling, referred to in Article 22(1) and (4), and at least in those cases give meaningful information about the logic involved and the significance and envisaged consequences. gdpr.automated-decision : information_digest Regulation (EU) 2016/679, Article 22(1) A person has the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning them or similarly significantly affects them. gdpr.automated-decision : subject_ref, decision_ref, solely_automated, significant_effect Regulation (EU) 2016/679, Article 22(2) That does not apply where the decision is necessary for entering into or performing a contract, is authorised by Union or Member State law with suitable safeguards, or is based on the person's explicit consent. gdpr.automated-decision : exception, basis_ref Regulation (EU) 2016/679, Article 22(3) Where the contract or explicit-consent exception applies, the controller implements suitable measures, at least the right to obtain human intervention, to express one's point of view and to contest the decision. gdpr.automated-decision : safeguards_offered gdpr.human-intervention : subject_ref, decision_record_id, request, requested_at, reviewer_ref, result, reasons, concluded_at gdpr.automated-decision Automated decision taken When: When a decision falling within Article 22(1) is taken, or when you conclude a decision does not fall within it. Field What to put in it Type Mapped to subject_ref Your pseudonymous reference for the person. Never a name, an email or a national number. your reference Article 22(1) decision_ref Your reference for the decision. your reference Article 22(1) solely_automated Whether the decision was based solely on automated processing. true or false Article 22(1) significant_effect Whether it produces legal effects concerning the person or similarly significantly affects them. true or false Article 22(1) exception The Article 22(2) exception relied on, or that the decision is outside Article 22(1). one of contract , union-or-member-state-law , explicit-consent , not-within-article-22 Article 22(2) basis_ref (optional) Your reference for the contract, the legal provision or the consent record relied on. your reference Article 22(2) Article 5(2) information_digest (optional) SHA-256 of the notice given to the person about the automated decision, as given. SHA-256 Article 13(2)(f) and Article 14(2)(g) safeguards_offered The safeguards offered with the decision. list of human-intervention , express-point-of-view , contest-decision Article 22(3) outcome The decision, in a short phrase. one line Article 5(2) decided_at When the decision was taken, UTC. UTC time Article 5(2) gdpr.human-intervention Human intervention requested and handled When: When the person asks for human intervention, expresses a point of view or contests the decision, and again when that request is concluded. Field What to put in it Type Mapped to subject_ref Your pseudonymous reference for the person. Never a name. your reference Article 22(3) decision_record_id The record_id of the gdpr.automated-decision record concerned. your reference Article 22(3) request What the person asked for. one of human-intervention , express-point-of-view , contest-decision Article 22(3) requested_at When the request arrived, UTC. UTC time Article 22(3) reviewer_ref (optional) Your reference for the person who reviewed. Never a name. your reference Article 22(3) result Where the request stands, or how it ended. one of pending , upheld , changed , reversed Article 22(3) reasons (optional) Why the decision was upheld, changed or reversed. text Article 22(3) Article 5(2) concluded_at (optional) When the review concluded, UTC. UTC time Article 22(3) Recording that a decision was outside Article 22(1) is as useful as recording one inside it: it shows the question was asked at the time. Insurance Distribution Directive pack idd version 1 Insurance distribution: demands and needs, advice and product information What an insurance distributor records to show, later, what the customer needed, what was proposed and why, and what information the customer received before the contract. The law Directive (EU) 2016/97 (Insurance Distribution Directive) (CELEX 32016L0097) Kinds idd.demands-needs demands and needs idd.recommendation personalised recommendation idd.ipid-provided product information document handed over idd.pre-contract-disclosure pre-contract information provided idd.product-version product version approved The file /packs/idd.json , with an example record for every kind How long to keep it This pack does not state how long to keep these records. Check the period that applies to you with your supervisor or your professional federation. In Belgium the directive is transposed by the Act of 4 April 2014 on insurance, and the FSMA supervises its conduct rules. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Directive (EU) 2016/97, Article 20(1), first subparagraph Before a contract is concluded, specify the customer's demands and needs on the basis of information obtained from the customer, and give objective information about the product in a comprehensible form. idd.demands-needs : customer_ref, product_class, information_obtained, demands_and_needs, specified_at Directive (EU) 2016/97, Article 20(1), second subparagraph Any contract proposed must be consistent with the customer's insurance demands and needs. idd.demands-needs : proposed_product_ref, consistency_note Directive (EU) 2016/97, Article 20(1), third subparagraph Where advice is given before a specific contract is concluded, give the customer a personalised recommendation explaining why a particular product would best meet their demands and needs. idd.demands-needs : advice_given idd.recommendation : customer_ref, demands_needs_record_id, recommended_product_ref, reasons, document_digest, given_at Directive (EU) 2016/97, Article 20, paragraphs 4 to 9 For non-life products, give the product information by way of the standardised insurance product information document (IPID). idd.ipid-provided : customer_ref, product_ref, ipid_version, ipid_digest, provided_at idd.pre-contract-disclosure : contract_ref, documents, provided_at Directive (EU) 2016/97, Article 23 How information reaches the customer: on paper, or on another durable medium or a website where the conditions for that are met. idd.recommendation : document_digest, medium idd.ipid-provided : medium idd.pre-contract-disclosure : customer_ref, contract_ref, documents, medium, provided_at, product_version_record_id Directive (EU) 2016/97, Article 25(1) Insurance undertakings, and intermediaries that manufacture insurance products, maintain, operate and review a process for the approval of each insurance product, or significant adaptation of an existing one, before it is marketed or distributed to customers. idd.product-version : product_ref, version, terms_digest, target_market_digest, approved_at, effective_from idd.demands-needs Demands and needs When: Once the customer's demands and needs are specified, before any contract is concluded. Field What to put in it Type Mapped to customer_ref Your own reference for the customer. Never a name. your reference Article 20(1), first subparagraph product_class The class of cover discussed, in your own terms, such as motor or home. one line Article 20(1), first subparagraph information_obtained What the customer told you, and how you obtained it. text Article 20(1), first subparagraph demands_and_needs The demands and needs you specified from that information. text Article 20(1), first subparagraph proposed_product_ref (optional) Your reference for the product you proposed, if any. your reference Article 20(1), second subparagraph consistency_note (optional) Why the proposed contract is consistent with those demands and needs. text Article 20(1), second subparagraph advice_given Whether you gave advice. If true, an idd.recommendation record follows. true or false Article 20(1), third subparagraph adviser_ref (optional) Your reference for the person who handled the customer. Never a name. your reference specified_at When the demands and needs were specified, UTC. UTC time Article 20(1), first subparagraph idd.recommendation Personalised recommendation When: When advice is given, once the recommendation is handed to the customer. Field What to put in it Type Mapped to customer_ref Your own reference for the customer. Never a name. your reference Article 20(1), third subparagraph demands_needs_record_id The record_id of the idd.demands-needs record this recommendation answers. your reference Article 20(1), third subparagraph recommended_product_ref Your reference for the product recommended. your reference Article 20(1), third subparagraph reasons Why this product would best meet the customer's demands and needs. text Article 20(1), third subparagraph document_digest (optional) SHA-256 of the recommendation document exactly as the customer received it. SHA-256 Article 20(1), third subparagraph Article 23 medium How the recommendation reached the customer. one of paper , durable-medium , website Article 23 given_at When the recommendation was given, UTC. UTC time Article 20(1), third subparagraph idd.ipid-provided Product information document handed over When: For a non-life product, when the IPID is given to the customer. Field What to put in it Type Mapped to customer_ref Your own reference for the customer. Never a name. your reference Article 20, paragraphs 4 to 9 product_ref Your reference for the product. your reference Article 20, paragraphs 4 to 9 ipid_version The version of the IPID, as the manufacturer labels it. one line Article 20, paragraphs 4 to 9 ipid_digest SHA-256 of the IPID file exactly as handed over. SHA-256 Article 20, paragraphs 4 to 9 medium How the IPID reached the customer. one of paper , durable-medium , website Article 23 provided_at When it was provided, UTC. UTC time Article 20, paragraphs 4 to 9 idd.pre-contract-disclosure Pre-contract information provided When: When the pre-contract documents reach the customer: each is sealed by its digest, and the customer can be given the receipt too, so both sides hold the same evidence of what was provided. Field What to put in it Type Mapped to customer_ref Your reference for the customer, never a name. your reference Article 23 contract_ref Your reference for the proposed contract. your reference Article 23 Article 20, paragraphs 4 to 9 documents SHA-256 of each document provided, exactly as provided (the IPID, the terms, the demands-and-needs statement). list of SHA-256 Article 23 Article 20, paragraphs 4 to 9 medium How the documents were provided. one of paper , durable-medium , website Article 23 provided_at When they were provided, UTC. UTC time Article 23 Article 20, paragraphs 4 to 9 product_version_record_id (optional) The record_id of the idd.product-version in force. your reference Article 23 idd.product-version Product version approved When: Each time a product, or a significant adaptation of it, is approved: its terms and target market are sealed by their digests, so which terms applied on a given day can be shown. Field What to put in it Type Mapped to product_ref Your reference for the product. your reference Article 25(1) version Your version label. one line Article 25(1) terms_digest SHA-256 of the terms and conditions exactly as approved. SHA-256 Article 25(1) target_market_digest (optional) SHA-256 of the target market definition as approved. SHA-256 Article 25(1) approved_at When the product version was approved, UTC. UTC time Article 25(1) effective_from From when it may be distributed. date Article 25(1) The Worker's own quote ledger (v1) can record what was proposed and priced; this pack covers what stays with you. EU Medical Device Regulations (MDR and IVDR) pack medical-devices version 1 Medical devices: technical documentation and incident reports What a manufacturer of a medical device or an in vitro diagnostic records so that each version of its technical documentation, its declaration of conformity and its certificates, and each serious incident or field safety corrective action report, can be shown later to be exactly what existed at the time. The law Regulation (EU) 2017/745 (Medical Device Regulation) (CELEX 32017R0745) Regulation (EU) 2017/746 (In Vitro Diagnostic Medical Devices Regulation) (CELEX 32017R0746) Kinds device.technical-documentation technical documentation, declaration or certificate version released device.incident-report serious incident or field safety corrective action reported The file /packs/medical-devices.json , with an example record for every kind How long to keep it Article 10(8) of the MDR sets at least 10 years after the last device covered by the declaration of conformity has been placed on the market, and at least 15 years for implantable devices; Article 10(7) of the IVDR sets at least 10 years. This pack does not decide your period. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Regulation (EU) 2017/745, Article 10(4) Manufacturers of devices other than custom-made devices draw up and keep up to date technical documentation that allows the conformity of the device to be assessed, including the elements set out in Annexes II and III. device.technical-documentation : device_ref, document_type, document_version, document_digest, released_at, previous_version_record_id Regulation (EU) 2017/745, Article 10(8) Manufacturers keep the technical documentation, the EU declaration of conformity and any relevant certificate, including amendments and supplements, available to the competent authorities for at least 10 years after the last device covered by the declaration has been placed on the market, and for implantable devices at least 15 years. device.technical-documentation : regulation, document_type, document_digest, implantable, retain_until Regulation (EU) 2017/745, Article 87(1) Manufacturers report serious incidents and field safety corrective actions concerning devices made available on the Union market to the competent authorities, through the electronic system referred to in Article 92. device.incident-report : device_ref, regulation, report_ref, subject, submitted_at, report_digest Regulation (EU) 2017/745, Article 87(3) A serious incident is reported immediately after the manufacturer has established, or finds reasonably possible, a causal relationship with its device, and not later than 15 days after becoming aware of it. device.incident-report : aware_at, submitted_at Regulation (EU) 2017/745, Article 87(6) Where necessary to report in time, the manufacturer may submit an initial report that is incomplete, followed by a complete report. device.incident-report : report_type Regulation (EU) 2017/746, Article 10(4) Manufacturers draw up and keep up to date technical documentation that allows the conformity of the device to be assessed, including the elements set out in Annexes II and III. device.technical-documentation : device_ref, document_type, document_version, document_digest, released_at, previous_version_record_id Regulation (EU) 2017/746, Article 10(7) Manufacturers keep the technical documentation, the EU declaration of conformity and any relevant certificate, including amendments and supplements, available to the competent authorities for at least 10 years after the last device covered by the declaration has been placed on the market. device.technical-documentation : regulation, document_type, document_digest, retain_until Regulation (EU) 2017/746, Article 82(1) Manufacturers report serious incidents and field safety corrective actions concerning devices made available on the Union market to the competent authorities, through the electronic system referred to in Article 87. device.incident-report : device_ref, regulation, report_ref, subject, submitted_at, report_digest Regulation (EU) 2017/746, Article 82(3) A serious incident is reported immediately after the manufacturer has established, or finds reasonably possible, a causal relationship with its device, and not later than 15 days after becoming aware of it. device.incident-report : aware_at, submitted_at device.technical-documentation Technical documentation, declaration or certificate version released When: Each time a version of the technical documentation, the EU declaration of conformity or a certificate, or an amendment or supplement to one, is released. Field What to put in it Type Mapped to device_ref Your reference for the device, or its Basic UDI-DI. your reference Article 10(4) Article 10(4) regulation Which regulation the device falls under. one of mdr , ivdr Article 10(8) Article 10(7) document_type What was released. one of technical-documentation , declaration-of-conformity , certificate Article 10(4) Article 10(8) Article 10(4) Article 10(7) document_version The version, as written on it. one line Article 10(4) Article 10(4) document_digest SHA-256 of the file exactly as released, or of the file list of a documentation folder, one line per file with its own SHA-256. SHA-256 Article 10(4) Article 10(8) Article 10(4) Article 10(7) released_at When it was released, UTC. UTC time Article 10(4) Article 10(4) implantable (optional) Whether the device is implantable, for which the MDR's period is at least 15 years. true or false Article 10(8) previous_version_record_id (optional) The record_id of the version this one replaces; the earlier version stays sealed. your reference Article 10(4) Article 10(4) retain_until (optional) The date until which you have decided to keep it. date Article 10(8) Article 10(7) device.incident-report Serious incident or field safety corrective action reported When: Each time a report is submitted to the competent authority: the initial report, each follow-up and the final report. Field What to put in it Type Mapped to device_ref Your reference for the device, or its Basic UDI-DI. your reference Article 87(1) Article 82(1) regulation Which regulation the device falls under. one of mdr , ivdr Article 87(1) Article 82(1) report_ref Your reference for the report; never a patient's name or a direct identifier. your reference Article 87(1) Article 82(1) subject What is reported. one of serious-incident , field-safety-corrective-action Article 87(1) Article 82(1) report_type Which report this is. one of initial , follow-up , final Article 87(6) aware_at When you became aware of the incident, UTC; the reporting periods run from here. UTC time Article 87(3) Article 82(3) submitted_at When the report was submitted, UTC. UTC time Article 87(1) Article 87(3) Article 82(1) Article 82(3) report_digest SHA-256 of the report exactly as submitted. SHA-256 Article 87(1) Article 82(1) Only digests are sent; the documentation and the reports stay with you, as does anything about a patient. A new version is a new record that names the one it replaces, and the earlier one stays sealed. Whether a device is in scope, in which class, and whether software that is a medical device is also a high-risk AI system under Article 6(1) of the AI Act, is for the manufacturer to establish; for such a system's logs, use the ai-act pack beside this one. MiFID II pack mifid2 version 1 Investment advice: the suitability assessment and the statement on suitability What an investment firm records to show what it knew about the client when it advised, what it advised, and that the statement on suitability existed, unaltered, before the transaction. The law Directive 2014/65/EU (MiFID II) (CELEX 32014L0065) Commission Delegated Regulation (EU) 2017/565 (CELEX 32017R0565) Kinds mifid.suitability suitability assessment mifid.suitability-statement statement on suitability mifid.call-recording conversation recorded mifid.sustainability-reasons sustainability preferences not met The file /packs/mifid2.json , with an example record for every kind How long to keep it Article 16(7) states five years for the recordings and communications it covers, and up to seven years where the competent authority requests it. For the other records this pack covers, check the period with your competent authority; in Belgium the FSMA supervises these conduct rules. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Directive 2014/65/EU, Article 25(2) When providing investment advice or portfolio management, obtain the information needed on the client's knowledge and experience in the relevant investment field, financial situation including ability to bear losses, and investment objectives including risk tolerance, so as to recommend what is suitable. mifid.suitability : client_ref, service, knowledge_and_experience, financial_situation, ability_to_bear_losses, investment_objectives, risk_tolerance, assessed_at Directive 2014/65/EU, Article 25(6) When providing investment advice, before the transaction is made, give the retail client a statement on suitability in a durable medium, specifying the advice given and how it meets the client's preferences, objectives and other characteristics. mifid.suitability-statement : client_ref, assessment_record_id, instruments, advice, how_it_meets_the_client, statement_digest, provided_at Commission Delegated Regulation (EU) 2017/565, Article 54 How the suitability assessment is carried out, and what the suitability report must contain. mifid.suitability : knowledge_and_experience, financial_situation, investment_objectives, other_preferences, method_ref mifid.suitability-statement : how_it_meets_the_client Commission Delegated Regulation (EU) 2017/565, Article 54(10), second and third subparagraphs A firm does not recommend or trade financial instruments as meeting a client's sustainability preferences when they do not meet them; it explains the reasons to the client and keeps records of those reasons. Where no financial instrument meets the client's sustainability preferences and the client decides to adapt them, the firm keeps records of the client's decision, including the reasons for it. mifid.sustainability-reasons : client_ref, assessment_record_id, situation, reasons, preferences_stated, preferences_adapted, recorded_at Commission Delegated Regulation (EU) 2017/565, Article 72(1) Records are kept so the competent authority can access them readily and reconstitute each key stage, so any corrections or amendments and the contents before them can be easily ascertained, and so they cannot otherwise be manipulated or altered. mifid.suitability : assessed_at mifid.suitability-statement : assessment_record_id, transaction_ref mifid.call-recording : recording_digest Directive 2014/65/EU, Article 16(7) Records include the recording of telephone conversations and electronic communications relating at least to transactions concluded when dealing on own account and to client order services, including those that do not result in a transaction; they are provided to the client on request and kept for five years and, where the competent authority requests it, up to seven years. mifid.call-recording : client_ref, channel, relates_to, started_at, ended_at, recording_digest, recording_format, order_refs mifid.suitability Suitability assessment When: When the information for the suitability assessment has been obtained and assessed, before advice is given. Field What to put in it Type Mapped to client_ref Your own reference for the client. Never a name. your reference Article 25(2) service The service the assessment is for. one of investment-advice , portfolio-management Article 25(2) knowledge_and_experience The client's knowledge and experience in the investment field relevant to the product or service. text Article 25(2) Article 54 financial_situation The client's financial situation, as obtained. text Article 25(2) Article 54 ability_to_bear_losses Your assessment of the client's ability to bear losses. one line Article 25(2) investment_objectives The client's investment objectives. text Article 25(2) Article 54 risk_tolerance The client's risk tolerance, in your own scale. one line Article 25(2) other_preferences (optional) Any other preference your assessment method records. text Article 54 method_ref (optional) Your reference for the questionnaire or method version used. your reference Article 54 adviser_ref (optional) Your reference for the adviser. Never a name. your reference assessed_at When the assessment was completed, UTC. UTC time Article 25(2) Article 72(1) mifid.suitability-statement Statement on suitability When: When the statement on suitability is given to the retail client, before the transaction is made. Field What to put in it Type Mapped to client_ref Your own reference for the client. Never a name. your reference Article 25(6) assessment_record_id The record_id of the mifid.suitability record the advice rests on. your reference Article 25(6) Article 72(1) instruments The instruments advised, by identifier such as ISIN. list of your reference Article 25(6) advice The advice given. text Article 25(6) how_it_meets_the_client How the advice meets the client's preferences, objectives and other characteristics. text Article 25(6) Article 54 statement_digest SHA-256 of the statement exactly as the client received it. SHA-256 Article 25(6) provided_at When the statement was provided, UTC. The log's own time bounds it from above. UTC time Article 25(6) transaction_ref (optional) Your reference for the transaction that followed, if any. your reference Article 72(1) mifid.call-recording Conversation recorded When: When a recorded telephone conversation or electronic communication ends: the recording is sealed by its digest at once, so it can be shown to be unedited whenever the client or the supervisor asks for it. Field What to put in it Type Mapped to client_ref Your reference for the client, never a name. your reference Article 16(7) channel The kind of communication. one of telephone , video , chat , email , other Article 16(7) relates_to What the communication relates to. one of own-account-dealing , reception-transmission , execution , advice , other Article 16(7) started_at When it started, UTC. UTC time Article 16(7) ended_at When it ended, UTC. UTC time Article 16(7) recording_digest SHA-256 of the recording file exactly as stored. SHA-256 Article 16(7) Article 72(1) recording_format (optional) The file format, such as wav or opus. one line Article 16(7) order_refs (optional) Your references for the orders or transactions it concerns. list of your reference Article 16(7) mifid.sustainability-reasons Sustainability preferences not met When: When no financial instrument meets the sustainability preferences the client stated: when you explain why you do not recommend one as meeting them, or when the client decides to adapt them. Field What to put in it Type Mapped to client_ref Your own reference for the client. Never a name. your reference Article 54(10), second and third subparagraphs assessment_record_id The record_id of the mifid.suitability record in which the preferences were stated. your reference Article 54(10), second and third subparagraphs situation Which record this is: the reasons for not recommending an instrument as meeting the client's sustainability preferences, or the client's decision to adapt them. one of not-recommended-as-meeting , client-adapted-preferences Article 54(10), second and third subparagraphs reasons The reasons explained to the client, or the reasons for the client's decision. text Article 54(10), second and third subparagraphs preferences_stated (optional) The sustainability preferences as the client stated them. text Article 54(10), second and third subparagraphs preferences_adapted (optional) The preferences as the client adapted them, when they did. text Article 54(10), second and third subparagraphs adviser_ref (optional) Your reference for the adviser. Never a name. your reference recorded_at When the explanation or the client's decision was recorded, UTC. UTC time Article 54(10), second and third subparagraphs A correction is a new record whose supersedes field names the record it corrects; the earlier record stays in the log. That is how Article 72(1)'s requirement that corrections and earlier contents can be ascertained shows up in the evidence. NIS2 and data breaches pack nis2 version 1 Security incidents and logs: NIS2 incident handling and reporting, and personal-data breaches What an organisation records so that, after a security incident, it can show which logs existed before the incident, what it knew and when, what it did, and what it reported to whom. The law Directive (EU) 2022/2555 (NIS2 Directive) (CELEX 32022L2555) Regulation (EU) 2016/679 (General Data Protection Regulation) (CELEX 32016R0679) Kinds security.log-segment log segment sealed security.incident incident recorded at a stage security.notification report or notification sent The file /packs/nis2.json , with an example record for every kind How long to keep it Neither NIS2 nor GDPR Article 33(5) states how long to keep these records, and this pack does not decide it for you. Check the period with your competent authority or supervisory authority. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Directive (EU) 2022/2555, Article 21(1) Essential and important entities take appropriate and proportionate technical, operational and organisational measures to manage the risks posed to the security of the network and information systems they use. security.log-segment : source_ref, log_digest Directive (EU) 2022/2555, Article 21(2)(b) Those measures include incident handling. security.log-segment : source_ref, segment_index, segment_start, segment_end, line_count, log_digest, previous_segment_record_id security.incident : incident_ref, stage, description, actions, stage_at, log_segment_record_ids Directive (EU) 2022/2555, Article 23(1) Notify the CSIRT or, where applicable, the competent authority without undue delay of any significant incident, and where appropriate the recipients of the services concerned. security.incident : significant security.notification : recipient Directive (EU) 2022/2555, Article 23(4) For a significant incident: an early warning within 24 hours of becoming aware of it, an incident notification within 72 hours, an intermediate report on request, and a final report within one month of the incident notification. security.incident : aware_at security.notification : incident_ref, stage, aware_at, submitted_at, report_digest Regulation (EU) 2016/679, Article 33(1) Notify a personal data breach to the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of it, unless it is unlikely to result in a risk to people's rights and freedoms; a later notification gives the reasons for the delay. security.incident : aware_at security.notification : incident_ref, stage, recipient, aware_at, submitted_at, report_digest, delay_reason Regulation (EU) 2016/679, Article 33(5) Document every personal data breach, comprising the facts relating to it, its effects and the remedial action taken, so the supervisory authority can verify compliance with the article. security.log-segment : segment_start, segment_end, log_digest security.incident : incident_ref, personal_data_breach, description, effects, actions, log_segment_record_ids security.notification : report_digest Regulation (EU) 2016/679, Article 34(1) When a breach is likely to result in a high risk to people's rights and freedoms, communicate it to the people concerned without undue delay. security.notification : stage, recipient security.log-segment Log segment sealed When: Each time a segment of a system's log is closed: after a number of lines, a number of seconds, or a file rotation. The logseal tool in the recorder kit does this for you. Field What to put in it Type Mapped to source_ref Your reference for the system or log source, such as fw-edge-1 or auth-server. your reference Article 21(1) Article 21(2)(b) segment_index The position of this segment in its source's sequence, from 0, so a missing segment is caught. whole number Article 21(2)(b) segment_start When the segment was opened, UTC. UTC time Article 21(2)(b) Article 33(5) segment_end When the segment was closed, UTC. UTC time Article 21(2)(b) Article 33(5) line_count How many lines the segment holds, so a shortened segment is caught. whole number Article 21(2)(b) log_digest SHA-256 of the segment exactly as archived. SHA-256 Article 21(1) Article 21(2)(b) Article 33(5) log_format (optional) The format of the lines, such as syslog, journald-json or cloudtrail-json. one line previous_segment_record_id (optional) The record_id of the segment before this one, so a removed segment is caught. your reference Article 21(2)(b) security.incident Incident recorded at a stage When: At each stage of a security incident: detected, assessed, contained, resolved, closed. Field What to put in it Type Mapped to incident_ref Your reference for the incident. your reference Article 21(2)(b) Article 33(5) stage The stage this record captures. one of detected , assessed , contained , resolved , closed Article 21(2)(b) significant Whether you consider it a significant incident under NIS2 at this stage. one of yes , no , not-yet-assessed Article 23(1) personal_data_breach Whether it is a personal data breach, as known at this stage. one of yes , no , not-yet-known Article 33(5) description The facts, as known at this stage. Refer to people by your own references, never by name. text Article 21(2)(b) Article 33(5) effects (optional) Its effects, as known at this stage. text Article 33(5) actions (optional) The remedial action taken, and what will be done to prevent recurrence. text Article 21(2)(b) Article 33(5) aware_at When you became aware of the incident, UTC: the reporting deadlines run from here. UTC time Article 23(4) Article 33(1) stage_at When this stage was reached, UTC. UTC time Article 21(2)(b) log_segment_record_ids (optional) The record_ids of the sealed log segments that cover the incident. list of your reference Article 21(2)(b) Article 33(5) security.notification Report or notification sent When: Each time something is sent about an incident: an early warning, a notification, an intermediate or final report, a breach notification, a communication to the people concerned. Field What to put in it Type Mapped to incident_ref Your reference for the incident. your reference Article 23(4) Article 33(1) stage What was sent. one of early-warning , incident-notification , intermediate-report , final-report , breach-notification , data-subject-communication Article 23(4) Article 33(1) Article 34(1) recipient To whom it was sent. one of csirt , competent-authority , supervisory-authority , service-recipients , data-subjects Article 23(1) Article 33(1) Article 34(1) aware_at When you became aware of the incident, UTC, as stated in what was sent. UTC time Article 23(4) Article 33(1) submitted_at When it was sent, UTC. UTC time Article 23(4) Article 33(1) report_digest SHA-256 of what was sent, exactly as sent, so the copy you keep can be shown to be the one you sent. SHA-256 Article 23(4) Article 33(1) Article 33(5) delay_reason (optional) Why it was sent after the deadline, if it was. text Article 33(1) The logs never leave you: the logseal tool in the recorder kit archives each segment on your own storage and sends only the digest of its record. A segment sealed before an intruder controlled the machine that writes it cannot be changed afterwards without it showing; what that machine writes once it is controlled can be forged at the source, and no seal changes that. Financial entities under DORA report ICT incidents under DORA, whose own pack covers them. Public procurement pack procurement version 1 Public procurement: tenders received and evaluations fixed before prices are opened What a contracting authority records so it can show which tenders it received and when, and that the quality scores were fixed before the prices were opened: the documentation that has to justify each decision in a procedure. The law Directive 2014/24/EU (public procurement) (CELEX 32014L0024) Kinds procurement.tender-received tender received procurement.evaluation-freeze evaluation fixed at a stage The file /packs/procurement.json , with an example record for every kind How long to keep it Article 84(2) states at least three years from the date of award of the contract. Your national law or funding rules may set a longer period; this pack does not decide it for you. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Directive 2014/24/EU, Article 67(1) Contracting authorities base the award of public contracts on the most economically advantageous tender. procurement.evaluation-freeze : stage, criteria_digest Directive 2014/24/EU, Article 84(2) Contracting authorities document the progress of all procurement procedures and keep sufficient documentation to justify decisions taken at all stages, such as on communications with economic operators, internal deliberations, selection and award; the documentation is kept for at least three years from the date of award. procurement.tender-received : procedure_ref, lot_ref, tenderer_ref, received_at, tender_digest procurement.evaluation-freeze : procedure_ref, lot_ref, stage, frozen_at, criteria_digest, scores_digest, committee_refs procurement.tender-received Tender received When: When each tender arrives, before anything is opened: its digest is sealed, so a tender cannot be swapped or backdated. Field What to put in it Type Mapped to procedure_ref Your reference for the procedure. your reference Article 84(2) lot_ref (optional) The lot, if the procedure has lots. your reference Article 84(2) tenderer_ref Your reference for the tenderer. your reference Article 84(2) received_at When it was received, UTC. UTC time Article 84(2) tender_digest SHA-256 of the tender exactly as received (the archive the platform gives you). SHA-256 Article 84(2) procurement.evaluation-freeze Evaluation fixed at a stage When: When the committee fixes its scores at a stage, before the next stage begins: in particular, the quality scores before the price envelopes are opened. Field What to put in it Type Mapped to procedure_ref Your reference for the procedure. your reference Article 84(2) lot_ref (optional) The lot, if the procedure has lots. your reference Article 84(2) stage The stage the scores belong to. one of selection , quality-before-price , award Article 84(2) Article 67(1) frozen_at When the scores were fixed, UTC. UTC time Article 84(2) criteria_digest SHA-256 of the award criteria and weightings as published. SHA-256 Article 67(1) Article 84(2) scores_digest SHA-256 of the score sheet exactly as fixed. SHA-256 Article 84(2) committee_refs Your references for the committee members who fixed it. list of your reference Article 84(2) The tenders and the score sheets stay with you; only their digests are sealed. Utilities procurement under Directive 2014/25/EU and concessions under Directive 2014/23/EU have their own documentation articles, not mapped here. VAT Directive invoices pack vat-invoice version 1 Invoices: authenticity of origin and integrity of content from issue to the end of storage What a business records so it can show that an invoice it issued or received has not been altered since the day it was issued, without sending the invoice anywhere. One business control among those the directive leaves to the taxable person. The law Council Directive 2006/112/EC (VAT Directive) (CELEX 32006L0112) Kinds invoice.issued invoice issued invoice.received invoice received The file /packs/vat-invoice.json , with an example record for every kind How long to keep it Article 247(1) leaves the storage period to each Member State. This pack does not decide it for you; check the period where the supply takes place. What each article asks, in our words, and where a record answers it Article What it asks (our summary; the linked text is the law) Recorded in Council Directive 2006/112/EC, Article 233(1) The authenticity of the origin, the integrity of the content and the legibility of an invoice, on paper or electronic, are ensured from the point in time of issue until the end of the period for storage; each taxable person determines how, for example by business controls that create a reliable audit trail between an invoice and a supply. invoice.issued : invoice_ref, issued_on, invoice_digest, invoice_format, supply_ref invoice.received : supplier_ref, invoice_digest, invoice_format Council Directive 2006/112/EC, Article 244 Every taxable person ensures that copies of the invoices issued by it, by its customer or by a third party on its behalf, and all the invoices it has received, are stored. invoice.issued : invoice_ref, invoice_digest, customer_ref invoice.received : invoice_ref, supplier_ref, received_at, invoice_digest invoice.issued Invoice issued When: When the invoice is issued, before it is sent: the file exactly as issued is sealed by its digest. Field What to put in it Type Mapped to invoice_ref The invoice number. your reference Article 233(1) Article 244 issued_on The date of issue on the invoice. date Article 233(1) invoice_digest SHA-256 of the invoice file exactly as issued (the XML of an e-invoice, or the PDF). SHA-256 Article 233(1) Article 244 invoice_format The form of the invoice. one of peppol-bis-billing-3 , en16931-ubl , en16931-cii , pdf , paper-scan , other Article 233(1) customer_ref (optional) Your reference for the customer. your reference Article 244 supply_ref (optional) Your reference for the supply invoiced (an order or a delivery), the link in the audit trail. your reference Article 233(1) invoice.received Invoice received When: When an invoice is received, before it is processed. Field What to put in it Type Mapped to invoice_ref The supplier's invoice number. your reference Article 244 supplier_ref Your reference for the supplier. your reference Article 244 Article 233(1) received_at When it was received, UTC. UTC time Article 244 invoice_digest SHA-256 of the invoice file exactly as received. SHA-256 Article 233(1) Article 244 invoice_format The form of the invoice. one of peppol-bis-billing-3 , en16931-ubl , en16931-cii , pdf , paper-scan , other Article 233(1) The invoice never leaves you: only its digest is sealed. A seal shows the invoice has not changed since it was sealed; it does not show the invoice is correct or that VAT is due. Directive (EU) 2025/516 (ViDA) amends the invoicing rules from dates it sets; check the consolidated text. Read this before using a pack What a pack is not A pack is a data format and a reading of the law, not legal advice, and using one does not make a firm meet any obligation. The summaries of articles are ours, written for orientation; the text on EUR-Lex, linked from each pack, is the law, and national rules and supervisory guidance add to it. We do not decide whether a regime applies to you. We never see your records, handle your customers, or report to your supervisor on your behalf. The log accepts any kind, packed or not: a pack is a convention you may adopt, never a condition of using the API. The packs, like the specification , are published so that another implementation can read the same records without us. Something wrong in a summary or a mapping? Tell us , and the correction goes in the changelog . ## What EU law asks you to record, and prove · RiskRouter URL: https://riskrouter.eu/obligations Obligations what the law already asks for The law already asks for the record. It does not say how you will prove it. Six EU laws ask a regulated firm to do something, and later to show it did: specify a customer’s needs, explain a recommendation, keep an AI system’s logs, record an incident. Each lets you keep that record in your own systems. None says how you will show, years later, that the record you produce is the one that existed at the time. Every system that stores a record can also edit it, so on the day a record is disputed, that is the question that decides. This page sets out, for each law, what its articles ask, when the record gets tested, and what a record sealed in the evidence log adds. The summaries are ours and the linked EUR-Lex text is the law. It is orientation, not legal advice: whether an article applies to you, and from when, is yours to establish. Fourteen laws, one question Law When the record is tested What a sealed record adds Insurance Distribution Directive A complaint to the insurer or the ombudsman, a claim refused because the cover did not match the need, a supervisor's inspection. The demands-and-needs note and the recommendation are sealed when they are written, so the note you produce is shown to be the one that existed before the contract, not one written after the complaint. MiFID II A client disputes the advice after a loss; the supervisor asks you to reconstitute each stage of it. A correction is a new sealed record that names the record it corrects, and the earlier one stays, so what was changed and what stood before can be ascertained, as Article 72(1) describes. EU AI Act A person affected by a decision asks for an explanation; a market surveillance authority asks for the logs. Each log segment is sealed as it is written, with its event count and the segment before it, so a shortened, edited or missing segment shows. GDPR Article 22 A person contests an automated decision; the data protection authority asks you to demonstrate compliance. The basis of the decision, and any human review of it, are shown to have been recorded at the time, while the record, which holds personal data, stays with you and erasable. NIS2 and data breaches A CSIRT or regulator asks what happened and when you knew; a cyber insurer assesses a claim; a customer or a court disputes your account of a breach. Logs sealed segment by segment before the incident cannot be rewritten afterwards without it showing, and each report you sent is shown to be the one you hold, sent when you say. DORA Your authority reviews a major incident; an auditor compares the report you hold with the report you sent. The incident record and each report are sealed when they are made, so the report you hold today is shown to be the one you held on the day you submitted it. GDPR consent and erasure A person says they never consented, or that their data was not erased; the data protection authority asks you to demonstrate it. Each consent is sealed with the notice the person saw, and each erasure when it is done, while the personal data stays in your systems and erasable. Consumer Credit Directive A borrower in difficulty says the credit should never have been granted; a consumer asks for a human to review an automated assessment. The assessment is sealed with the digests of its inputs and its model before the agreement, so it can be shown to have come first, and each review is sealed when it is given. Anti-Money Laundering Regulation A supervisor reviews a customer onboarded years ago, or asks which sanctions list you screened against on a given day. The customer file is sealed by the digests of its documents when it is approved, and each sanctions list and screening when it happens, so no document can be swapped in later and the list version is fixed. Cyber Resilience Act A market surveillance authority asks for the technical documentation; a CSIRT or a customer asks when you knew of an exploited vulnerability and when you reported it. Each release is sealed with its software bill of materials, and each report when it is sent, so what shipped and when you reported are not your word alone. VAT Directive invoices A tax audit questions an invoice years later; a customer or supplier disputes what an invoice said. Each invoice is sealed by its digest when it is issued or received, a business control showing its content has not changed since, without the invoice leaving you. Public procurement A losing tenderer challenges the award; an auditor asks you to justify the decision at each stage. Each tender is sealed when it arrives and the quality scores before the prices are opened, so neither can be shown to have changed afterwards. EU Clinical Trials Regulation An inspection of the trial master file; a question about a safety report, years after the trial; an archive moved to new media or a new owner. Each essential document is sealed when it is filed and each new version names the one it replaces, so any alteration to the file is traceable, as Article 58 asks, and the archive is shown complete by its file list. EU Medical Device Regulations (MDR and IVDR) A notified body or a competent authority asks for the technical documentation that applied to a device on a given date; an incident report is questioned. Each version of the technical documentation, the declaration and the certificates is sealed when it is released, and each incident report when it is submitted, so the version that applied on a date, and when you reported, are not your word alone. Insurance Distribution Directive Who it concerns Insurance distributors: brokers, agents, and insurers selling directly. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A complaint to the insurer or the ombudsman, a claim refused because the cover did not match the need, a supervisor's inspection. What a sealed record adds The demands-and-needs note and the recommendation are sealed when they are written, so the note you produce is shown to be the one that existed before the contract, not one written after the complaint. How long to keep it This pack does not state how long to keep these records. Check the period that applies to you with your supervisor or your professional federation. In Belgium the directive is transposed by the Act of 4 April 2014 on insurance, and the FSMA supervises its conduct rules. What to record The Insurance Distribution Directive pack : idd.demands-needs , idd.recommendation , idd.ipid-provided , idd.pre-contract-disclosure , idd.product-version Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Directive (EU) 2016/97, Article 20(1), first subparagraph Before a contract is concluded, specify the customer's demands and needs on the basis of information obtained from the customer, and give objective information about the product in a comprehensible form. Directive (EU) 2016/97, Article 20(1), second subparagraph Any contract proposed must be consistent with the customer's insurance demands and needs. Directive (EU) 2016/97, Article 20(1), third subparagraph Where advice is given before a specific contract is concluded, give the customer a personalised recommendation explaining why a particular product would best meet their demands and needs. Directive (EU) 2016/97, Article 20, paragraphs 4 to 9 For non-life products, give the product information by way of the standardised insurance product information document (IPID). Directive (EU) 2016/97, Article 23 How information reaches the customer: on paper, or on another durable medium or a website where the conditions for that are met. Directive (EU) 2016/97, Article 25(1) Insurance undertakings, and intermediaries that manufacture insurance products, maintain, operate and review a process for the approval of each insurance product, or significant adaptation of an existing one, before it is marketed or distributed to customers. MiFID II Who it concerns Investment firms that give investment advice or manage portfolios. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A client disputes the advice after a loss; the supervisor asks you to reconstitute each stage of it. What a sealed record adds A correction is a new sealed record that names the record it corrects, and the earlier one stays, so what was changed and what stood before can be ascertained, as Article 72(1) describes. How long to keep it Article 16(7) states five years for the recordings and communications it covers, and up to seven years where the competent authority requests it. For the other records this pack covers, check the period with your competent authority; in Belgium the FSMA supervises these conduct rules. What to record The MiFID II pack : mifid.suitability , mifid.suitability-statement , mifid.call-recording , mifid.sustainability-reasons Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Directive 2014/65/EU, Article 25(2) When providing investment advice or portfolio management, obtain the information needed on the client's knowledge and experience in the relevant investment field, financial situation including ability to bear losses, and investment objectives including risk tolerance, so as to recommend what is suitable. Directive 2014/65/EU, Article 25(6) When providing investment advice, before the transaction is made, give the retail client a statement on suitability in a durable medium, specifying the advice given and how it meets the client's preferences, objectives and other characteristics. Commission Delegated Regulation (EU) 2017/565, Article 54 How the suitability assessment is carried out, and what the suitability report must contain. Commission Delegated Regulation (EU) 2017/565, Article 54(10), second and third subparagraphs A firm does not recommend or trade financial instruments as meeting a client's sustainability preferences when they do not meet them; it explains the reasons to the client and keeps records of those reasons. Where no financial instrument meets the client's sustainability preferences and the client decides to adapt them, the firm keeps records of the client's decision, including the reasons for it. Commission Delegated Regulation (EU) 2017/565, Article 72(1) Records are kept so the competent authority can access them readily and reconstitute each key stage, so any corrections or amendments and the contents before them can be easily ascertained, and so they cannot otherwise be manipulated or altered. Directive 2014/65/EU, Article 16(7) Records include the recording of telephone conversations and electronic communications relating at least to transactions concluded when dealing on own account and to client order services, including those that do not result in a transaction; they are provided to the client on request and kept for five years and, where the competent authority requests it, up to seven years. EU AI Act Who it concerns Providers and deployers of high-risk AI systems, which include risk assessment and pricing for natural persons in life and health insurance, and assessing the creditworthiness of natural persons (Annex III, point 5). Whether it applies to you, and from when, is yours to establish from the text. When it is tested A person affected by a decision asks for an explanation; a market surveillance authority asks for the logs. What a sealed record adds Each log segment is sealed as it is written, with its event count and the segment before it, so a shortened, edited or missing segment shows. How long to keep it Articles 19(1) and 26(6) set a minimum of six months for the logs they cover, unless other law provides otherwise; this pack does not decide your period. Whether your system is high-risk, and from when these obligations apply to it, is yours to establish from the Regulation. What to record The EU AI Act pack : ai.log-segment , ai.decision , ai.oversight , ai.model-provenance , ai.screening-criteria , ai.content-credential , ai.training-data Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2024/1689, Article 12(1) High-risk AI systems must technically allow for the automatic recording of events (logs) over the lifetime of the system. Regulation (EU) 2024/1689, Article 12(2) Logging must enable recording of events relevant for identifying situations that may result in the system presenting a risk or in a substantial modification, for post-market monitoring (Article 72), and for monitoring the system's operation (Article 26(5)). Regulation (EU) 2024/1689, Article 14(4)(d) The people overseeing the system must be able to decide, in any particular situation, not to use it or to disregard, override or reverse its output. Regulation (EU) 2024/1689, Article 19(1) Providers keep the logs their high-risk AI systems generate automatically, to the extent the logs are under their control, for a period appropriate to the system's intended purpose and of at least six months, unless Union or national law, in particular on personal data, provides otherwise. Regulation (EU) 2024/1689, Article 19(2) Providers that are financial institutions subject to internal-governance requirements under Union financial services law keep those logs as part of the documentation kept under that law. Regulation (EU) 2024/1689, Article 26(6) Deployers keep the logs the system generates automatically, to the extent the logs are under their control, for a period appropriate to the system's intended purpose and of at least six months, unless applicable law provides otherwise. Regulation (EU) 2024/1689, Article 86(1) A person affected by a decision a deployer takes on the basis of output from a high-risk AI system listed in Annex III (other than point 2), with legal or similarly significant adverse effects, may obtain clear and meaningful explanations of the role of the AI system in the decision and of the main elements of the decision taken. Regulation (EU) 2024/1689, Article 50(2) Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, as far as technically feasible. Regulation (EU) 2024/1689, Article 10(2) Training, validation and testing data sets of high-risk AI systems are subject to data governance and management practices appropriate to the intended purpose, concerning in particular the data collection processes and the origin of the data (and, for personal data, the original purpose of the collection) and the data-preparation operations such as annotation, labelling, cleaning, updating, enrichment and aggregation. Regulation (EU) 2024/1689, Article 53(1)(a) Providers of general-purpose AI models draw up and keep up to date the technical documentation of the model, including its training and testing process and the results of its evaluation, to provide it on request to the AI Office and the national competent authorities. Regulation (EU) 2024/1689, Article 53(1)(c) Providers of general-purpose AI models put in place a policy to comply with Union law on copyright and related rights, in particular to identify and comply with a reservation of rights expressed under Article 4(3) of Directive (EU) 2019/790. Regulation (EU) 2024/1689, Article 53(1)(d) Providers of general-purpose AI models draw up and make publicly available a sufficiently detailed summary about the content used for training the model, according to a template provided by the AI Office. GDPR Article 22 Who it concerns Any controller whose decisions about people are made, wholly or partly, by automated processing. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A person contests an automated decision; the data protection authority asks you to demonstrate compliance. What a sealed record adds The basis of the decision, and any human review of it, are shown to have been recorded at the time, while the record, which holds personal data, stays with you and erasable. How long to keep it This pack does not state how long to keep these records. The records are yours and hold personal data under your control: keep them no longer than your purposes need, and they stay erasable because they never leave you. Only a salted digest is in the log, and without the record and its salt nobody, including us, can link it to anyone. What to record The GDPR Article 22 pack : gdpr.automated-decision , gdpr.human-intervention Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2016/679, Article 5(2) The controller is responsible for, and must be able to demonstrate, compliance with the principles in Article 5(1). Regulation (EU) 2016/679, Article 13(2)(f) and Article 14(2)(g) Inform the person of the existence of automated decision-making, including profiling, referred to in Article 22(1) and (4), and at least in those cases give meaningful information about the logic involved and the significance and envisaged consequences. Regulation (EU) 2016/679, Article 22(1) A person has the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning them or similarly significantly affects them. Regulation (EU) 2016/679, Article 22(2) That does not apply where the decision is necessary for entering into or performing a contract, is authorised by Union or Member State law with suitable safeguards, or is based on the person's explicit consent. Regulation (EU) 2016/679, Article 22(3) Where the contract or explicit-consent exception applies, the controller implements suitable measures, at least the right to obtain human intervention, to express one's point of view and to contest the decision. NIS2 and data breaches Who it concerns Essential and important entities under NIS2: medium-sized and large organisations in the sectors of its Annexes I and II, and some others whatever their size (Article 2); for personal-data breaches, every controller under the GDPR. Financial entities report ICT incidents under DORA instead. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A CSIRT or regulator asks what happened and when you knew; a cyber insurer assesses a claim; a customer or a court disputes your account of a breach. What a sealed record adds Logs sealed segment by segment before the incident cannot be rewritten afterwards without it showing, and each report you sent is shown to be the one you hold, sent when you say. How long to keep it Neither NIS2 nor GDPR Article 33(5) states how long to keep these records, and this pack does not decide it for you. Check the period with your competent authority or supervisory authority. What to record The NIS2 and data breaches pack : security.log-segment , security.incident , security.notification Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Directive (EU) 2022/2555, Article 21(1) Essential and important entities take appropriate and proportionate technical, operational and organisational measures to manage the risks posed to the security of the network and information systems they use. Directive (EU) 2022/2555, Article 21(2)(b) Those measures include incident handling. Directive (EU) 2022/2555, Article 23(1) Notify the CSIRT or, where applicable, the competent authority without undue delay of any significant incident, and where appropriate the recipients of the services concerned. Directive (EU) 2022/2555, Article 23(4) For a significant incident: an early warning within 24 hours of becoming aware of it, an incident notification within 72 hours, an intermediate report on request, and a final report within one month of the incident notification. Regulation (EU) 2016/679, Article 33(1) Notify a personal data breach to the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of it, unless it is unlikely to result in a risk to people's rights and freedoms; a later notification gives the reasons for the delay. Regulation (EU) 2016/679, Article 33(5) Document every personal data breach, comprising the facts relating to it, its effects and the remedial action taken, so the supervisory authority can verify compliance with the article. Regulation (EU) 2016/679, Article 34(1) When a breach is likely to result in a high risk to people's rights and freedoms, communicate it to the people concerned without undue delay. DORA Who it concerns Financial entities within DORA's scope, including insurers and insurance intermediaries, other than intermediaries that are micro, small or medium-sized enterprises (Article 2(3)). Whether it applies to you, and from when, is yours to establish from the text. When it is tested Your authority reviews a major incident; an auditor compares the report you hold with the report you sent. What a sealed record adds The incident record and each report are sealed when they are made, so the report you hold today is shown to be the one you held on the day you submitted it. How long to keep it This pack does not state how long to keep these records. Check the period with your competent authority. What to record The DORA pack : dora.incident , dora.incident-report , dora.register-snapshot Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2022/2554, Article 17(1) Define, establish and implement an ICT-related incident management process to detect, manage and notify ICT-related incidents. Regulation (EU) 2022/2554, Article 17(2) Record all ICT-related incidents and significant cyber threats, with procedures for consistent monitoring, handling and follow-up so root causes are identified, documented and addressed. Regulation (EU) 2022/2554, Article 18 Classify ICT-related incidents and determine their impact against the criteria the article sets. Regulation (EU) 2022/2554, Article 19(4) For a major ICT-related incident, submit to the competent authority an initial notification, an intermediate report and a final report. Regulation (EU) 2022/2554, Article 28(3) Maintain and update a register of information on all contractual arrangements for ICT services provided by ICT third-party service providers. GDPR consent and erasure Who it concerns Any controller that relies on consent, or that erases personal data on request or when its period runs out. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A person says they never consented, or that their data was not erased; the data protection authority asks you to demonstrate it. What a sealed record adds Each consent is sealed with the notice the person saw, and each erasure when it is done, while the personal data stays in your systems and erasable. How long to keep it The GDPR states no fixed period for these records: keep them as long as you may need to demonstrate the consent or the erasure, and no longer. This pack does not decide it for you. What to record The GDPR consent and erasure pack : gdpr.consent , gdpr.erasure Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2016/679, Article 5(2) The controller is responsible for, and must be able to demonstrate, compliance with the principles in Article 5(1). Regulation (EU) 2016/679, Article 7(1) Where processing is based on consent, the controller must be able to demonstrate that the person consented to the processing of their personal data. Regulation (EU) 2016/679, Article 7(3) The person may withdraw consent at any time, withdrawing must be as easy as giving consent, and the person is informed of this before giving it. Regulation (EU) 2016/679, Article 17(1) The person has the right to obtain the erasure of personal data concerning them without undue delay, and the controller must erase it without undue delay, where one of the grounds the article lists applies. Consumer Credit Directive Who it concerns Creditors granting consumer credit within the directive's scope, from 20 November 2026 as transposed in each Member State. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A borrower in difficulty says the credit should never have been granted; a consumer asks for a human to review an automated assessment. What a sealed record adds The assessment is sealed with the digests of its inputs and its model before the agreement, so it can be shown to have come first, and each review is sealed when it is given. How long to keep it Article 18 does not state how long to keep these records, and this pack does not decide it for you. The directive applies from 20 November 2026 as transposed in each Member State; check your national law. What to record The Consumer Credit Directive pack : credit.creditworthiness-assessment , credit.human-intervention Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Directive (EU) 2023/2225, Article 18(1) Before concluding a credit agreement, the creditor carries out a thorough assessment of the consumer's creditworthiness, in the consumer's interest, to prevent irresponsible lending practices and over-indebtedness. Directive (EU) 2023/2225, Article 18(8) Where the assessment involves automated processing of personal data, the consumer has the right to request and obtain human intervention on the part of the creditor, an explanation of the assessment, and to express their point of view and request a review of the assessment and of the decision. Anti-Money Laundering Regulation Who it concerns Obliged entities under the Anti-Money Laundering Regulation, from 10 July 2027: credit and financial institutions, crypto-asset service providers, and the other entities it lists. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A supervisor reviews a customer onboarded years ago, or asks which sanctions list you screened against on a given day. What a sealed record adds The customer file is sealed by the digests of its documents when it is approved, and each sanctions list and screening when it happens, so no document can be swapped in later and the list version is fixed. How long to keep it Article 77 states five years from the end of the business relationship, or from the date of an occasional transaction, and says when that period may be extended. The regulation applies from 10 July 2027. This pack does not decide your period. What to record The Anti-Money Laundering Regulation pack : aml.onboarding-file , aml.list-version , aml.screening Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2024/1624, Article 20(1) The customer due diligence measures: among them, identifying the customer and the beneficial owners and verifying their identity, and understanding the purpose and intended nature of the business relationship. Regulation (EU) 2024/1624, Article 20(1), point (d) Verify whether the customer or the beneficial owners are subject to targeted financial sanctions and, for a legal entity, whether persons subject to them control it or hold more than 50 % of its proprietary rights or a majority interest in it. Regulation (EU) 2024/1624, Article 26 Regularly verify again whether the condition in Article 20(1), point (d), is met, as often as the exposure to the risks of non-implementation and evasion of targeted financial sanctions requires; credit and financial institutions also do so upon a new designation. Regulation (EU) 2024/1624, Article 77(1) Retain, among other things, a copy of the documents and information obtained in performing customer due diligence, the record of the assessment of suspicious transactions, and the supporting evidence and records of transactions. Cyber Resilience Act Who it concerns Manufacturers of products with digital elements made available in the EU: reporting since 11 September 2026, most other obligations from 11 December 2027. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A market surveillance authority asks for the technical documentation; a CSIRT or a customer asks when you knew of an exploited vulnerability and when you reported it. What a sealed record adds Each release is sealed with its software bill of materials, and each report when it is sent, so what shipped and when you reported are not your word alone. How long to keep it The regulation requires the technical documentation to be kept for at least ten years after the product is placed on the market, or for the support period if that is longer. Article 14 has applied since 11 September 2026; most other obligations apply from 11 December 2027. This pack does not decide your period. What to record The Cyber Resilience Act pack : cra.release , cra.vulnerability-report Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2024/2847, Article 13 Manufacturers place products with digital elements on the market only if they meet the essential cybersecurity requirements of Annex I, whose Part II includes identifying and documenting the product's vulnerabilities and components, among other things by drawing up a software bill of materials. Regulation (EU) 2024/2847, Article 31 The technical documentation, with the contents Annex VII sets out, is drawn up before the product is placed on the market and kept up to date. Regulation (EU) 2024/2847, Article 14(1) Notify any actively exploited vulnerability in the product that the manufacturer becomes aware of, simultaneously to the CSIRT designated as coordinator and to ENISA, via the single reporting platform. Regulation (EU) 2024/2847, Article 14(2) An early warning within 24 hours of becoming aware, a vulnerability notification within 72 hours, and a final report no later than 14 days after a corrective or mitigating measure is available. VAT Directive invoices Who it concerns Every taxable person that issues or receives invoices. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A tax audit questions an invoice years later; a customer or supplier disputes what an invoice said. What a sealed record adds Each invoice is sealed by its digest when it is issued or received, a business control showing its content has not changed since, without the invoice leaving you. How long to keep it Article 247(1) leaves the storage period to each Member State. This pack does not decide it for you; check the period where the supply takes place. What to record The VAT Directive invoices pack : invoice.issued , invoice.received Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Council Directive 2006/112/EC, Article 233(1) The authenticity of the origin, the integrity of the content and the legibility of an invoice, on paper or electronic, are ensured from the point in time of issue until the end of the period for storage; each taxable person determines how, for example by business controls that create a reliable audit trail between an invoice and a supply. Council Directive 2006/112/EC, Article 244 Every taxable person ensures that copies of the invoices issued by it, by its customer or by a third party on its behalf, and all the invoices it has received, are stored. Public procurement Who it concerns Contracting authorities awarding public contracts under Directive 2014/24/EU. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A losing tenderer challenges the award; an auditor asks you to justify the decision at each stage. What a sealed record adds Each tender is sealed when it arrives and the quality scores before the prices are opened, so neither can be shown to have changed afterwards. How long to keep it Article 84(2) states at least three years from the date of award of the contract. Your national law or funding rules may set a longer period; this pack does not decide it for you. What to record The Public procurement pack : procurement.tender-received , procurement.evaluation-freeze Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Directive 2014/24/EU, Article 67(1) Contracting authorities base the award of public contracts on the most economically advantageous tender. Directive 2014/24/EU, Article 84(2) Contracting authorities document the progress of all procurement procedures and keep sufficient documentation to justify decisions taken at all stages, such as on communications with economic operators, internal deliberations, selection and award; the documentation is kept for at least three years from the date of award. EU Clinical Trials Regulation Who it concerns Sponsors and investigators of clinical trials on medicinal products under Regulation (EU) No 536/2014, and the software that keeps their trial master files. Whether it applies to you, and from when, is yours to establish from the text. When it is tested An inspection of the trial master file; a question about a safety report, years after the trial; an archive moved to new media or a new owner. What a sealed record adds Each essential document is sealed when it is filed and each new version names the one it replaces, so any alteration to the file is traceable, as Article 58 asks, and the archive is shown complete by its file list. How long to keep it Article 58 sets at least 25 years after the end of the trial for the content of the clinical trial master file, unless other Union law requires longer; subjects' medical files follow national law. This pack does not decide your period. What to record The EU Clinical Trials Regulation pack : clinical.tmf-document , clinical.tmf-archive , clinical.adverse-event-report , clinical.serious-breach-notification Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) No 536/2014, Article 41(2) The investigator records and documents all adverse events, unless the protocol provides differently, and reports serious adverse events to the sponsor without undue delay and not later than 24 hours after becoming aware of them, unless the protocol provides that no immediate reporting is required, with a follow-up report where relevant. Regulation (EU) No 536/2014, Article 41(3) The sponsor keeps detailed records of all adverse events reported to it by the investigator. Regulation (EU) No 536/2014, Article 52(1) The sponsor notifies the Member States concerned, through the EU portal, of a serious breach of the Regulation or of the protocol version applicable at the time, without undue delay and not later than seven days after becoming aware of it. Regulation (EU) No 536/2014, Article 57 The sponsor and the investigator each keep a clinical trial master file that at all times contains the essential documents allowing the conduct of the trial and the quality of the data generated to be verified, readily available and directly accessible to the Member States on request. Regulation (EU) No 536/2014, Article 58, first paragraph Unless other Union law requires longer, the sponsor and the investigator archive the content of the clinical trial master file for at least 25 years after the end of the trial; the medical files of subjects are archived in accordance with national law. Regulation (EU) No 536/2014, Article 58, fifth paragraph The media used to archive the content of the clinical trial master file keep it complete and legible throughout the archiving period. Regulation (EU) No 536/2014, Article 58, sixth paragraph Any alteration to the content of the clinical trial master file is traceable. EU Medical Device Regulations (MDR and IVDR) Who it concerns Manufacturers of medical devices and in vitro diagnostics, including software that is itself a device. Whether it applies to you, and from when, is yours to establish from the text. When it is tested A notified body or a competent authority asks for the technical documentation that applied to a device on a given date; an incident report is questioned. What a sealed record adds Each version of the technical documentation, the declaration and the certificates is sealed when it is released, and each incident report when it is submitted, so the version that applied on a date, and when you reported, are not your word alone. How long to keep it Article 10(8) of the MDR sets at least 10 years after the last device covered by the declaration of conformity has been placed on the market, and at least 15 years for implantable devices; Article 10(7) of the IVDR sets at least 10 years. This pack does not decide your period. What to record The EU Medical Device Regulations (MDR and IVDR) pack : device.technical-documentation , device.incident-report Article (links to the text on EUR-Lex) What it asks (our summary; the linked text is the law) Regulation (EU) 2017/745, Article 10(4) Manufacturers of devices other than custom-made devices draw up and keep up to date technical documentation that allows the conformity of the device to be assessed, including the elements set out in Annexes II and III. Regulation (EU) 2017/745, Article 10(8) Manufacturers keep the technical documentation, the EU declaration of conformity and any relevant certificate, including amendments and supplements, available to the competent authorities for at least 10 years after the last device covered by the declaration has been placed on the market, and for implantable devices at least 15 years. Regulation (EU) 2017/745, Article 87(1) Manufacturers report serious incidents and field safety corrective actions concerning devices made available on the Union market to the competent authorities, through the electronic system referred to in Article 92. Regulation (EU) 2017/745, Article 87(3) A serious incident is reported immediately after the manufacturer has established, or finds reasonably possible, a causal relationship with its device, and not later than 15 days after becoming aware of it. Regulation (EU) 2017/745, Article 87(6) Where necessary to report in time, the manufacturer may submit an initial report that is incomplete, followed by a complete report. Regulation (EU) 2017/746, Article 10(4) Manufacturers draw up and keep up to date technical documentation that allows the conformity of the device to be assessed, including the elements set out in Annexes II and III. Regulation (EU) 2017/746, Article 10(7) Manufacturers keep the technical documentation, the EU declaration of conformity and any relevant certificate, including amendments and supplements, available to the competent authorities for at least 10 years after the last device covered by the declaration has been placed on the market. Regulation (EU) 2017/746, Article 82(1) Manufacturers report serious incidents and field safety corrective actions concerning devices made available on the Union market to the competent authorities, through the electronic system referred to in Article 87. Regulation (EU) 2017/746, Article 82(3) A serious incident is reported immediately after the manufacturer has established, or finds reasonably possible, a causal relationship with its device, and not later than 15 days after becoming aware of it. What no law requires, and what the log does not do No law names a product None of these laws requires RiskRouter or any other product. You can meet every duty on this page without us. What we offer is a way to show you met it that does not depend on anyone’s word, including ours. It proves when, not whether A sealed record is shown to have existed, unchanged, no later than a signed and anchored head. The log does not know what the record says, so it cannot make a record correct, complete or sufficient, and recording something does not make anyone compliant. The record stays with you Only a salted fingerprint of each record is sent. The record, and any personal data in it, never reach us, so it stays yours to keep and to erase. Data protection It outlives us Every proof checks offline with the open-source verifier , against keys and anchors that are published. Continuity To start, take a sandbox key and record a fingerprint in a few minutes, or read what to record, rule by rule . ## DORA Article 30 contract terms, clause by clause · RiskRouter URL: https://riskrouter.eu/dora For procurement, risk and DORA review DORA Article 30, clause by clause A financial entity may rely on an ICT provider only under a written contract that contains the provisions Regulation (EU) 2022/2554 (DORA) , Article 30, lists. Below is each of those provisions, the clause we offer for it, and what is in place today. Where something is not in place, it says so. These are proposed terms. RiskRouter has no legal entity registered yet, so no contract can be signed today ( facts for your register of information ). When one can, these are the terms we will put in it, in one written document you can keep. The summaries of the Regulation are ours and the linked text is the law. This is not legal advice: your counsel should check the clauses against your own policy, and whether the function we support is critical or important is your assessment. Article 30(2): in every contract (a) The services, and subcontracting Requires A clear and complete description of the functions and services, saying whether subcontracting is permitted and on what conditions. We offer The service is described in the order form: the evidence log ( /api/v2 ), the quote ledger and pricing engine ( /api/v1 ), with the API reference as the technical description at the date of signature. Subcontracting is permitted only to the providers listed on the data protection page . We give you at least 30 days’ written notice before adding or replacing one, and you may terminate without charge if you object. In place today The description and the list of subprocessors are published. (b) Where services are provided and data is kept Requires The regions or countries where the services are provided and data is processed and stored, and advance notice of any change. We offer Stored in the EU: the database is in Frankfurt, Germany. Requests are handled at the Cloudflare location nearest the caller, which may be outside the EEA, and nothing is stored there. An encrypted daily backup is kept on GitHub for 90 days, outside the EU. At least 30 days’ written notice before any of these changes. In place today As stated, and published on Trust . (c) Availability, authenticity, integrity and confidentiality of data Requires Provisions on the availability, authenticity, integrity and confidentiality of data, including personal data. We offer Integrity is the product: entries are append-only in the database itself, hash-chained, under signed heads with public timestamps and a Bitcoin anchor, and you can check all of it without us. Your records never reach us: only salted fingerprints are sent. Only the SHA-256 of an API key is stored. A data processing agreement under GDPR Article 28 is attached where any personal data is processed on your behalf. In place today Built, tested in CI, and described in the specification and on Security . (d) Access to your data, and its return, whatever happens to us Requires Access, recovery and return of your data in an easily accessible format if the provider becomes insolvent, is resolved or stops, or the contract ends. We offer You can export everything you recorded at any time, in an open JSON format, without a ticket or a fee, and it verifies without us. We give at least 90 days’ written notice before discontinuing the service, and the export stays available throughout that period and any transition period under (3)(f). In place today The export, the open format and the public verifier. What you keep (e) Service levels Requires Service level descriptions, including their updates and revisions. We offer The service levels agreed for your contract, in the order form. Every change to the API and to the rating matrix is listed in the changelog , with at least 90 days’ written notice before we remove or change anything you rely on; live availability is on the status page . In place today No service level is offered during the validation build. It is agreed in writing per pilot. (f) Help when an incident touches the service Requires Assistance when an ICT incident related to the service occurs, at no additional cost or at a cost fixed in advance. We offer Assistance at no additional cost: the facts, logs and timeline we hold about the incident, and our help in establishing what it affected. In place today Incident notes are kept in the repository, and the ledger’s own history is independently checkable. (g) Cooperation with your authorities Requires Full cooperation with your competent and resolution authorities, and the people they appoint. We offer Full cooperation, without charge. In place today A commitment; nothing is needed to begin it. (h) Ending the contract Requires Termination rights and minimum notice periods that meet your authorities’ expectations, including the grounds in Article 28(7). We offer You may end the contract at any time, for any reason, with no notice period and no termination fee, as well as on every ground in Article 28(7). We may end it for convenience only with at least 90 days’ written notice, and never without the transition period under (3)(f) if you ask for it. In place today Leaving needs nothing from us but revoking your key. The steps to leave (i) Security awareness and resilience training Requires The conditions on which the provider takes part in your ICT security awareness programmes and digital operational resilience training (Article 13(6)). We offer We take part when you ask, at no charge. In place today A commitment. Article 30(3): where the service supports a critical or important function Whether it does is your assessment. If it does, the contract also needs the following, and the first is not something we can offer yet. Tell us before a pilot starts, so it is settled in writing rather than discovered later. (a) Measurable service levels Requires Full service level descriptions with precise quantitative and qualitative performance targets, so you can monitor the service and act when a target is missed. In place today Not offered. During the validation build we publish no availability target and give no service credit. A contract for a critical or important function would need them agreed first. (b) Notice of anything that affects the service Requires Notice periods and reporting obligations, including notice of any development that might materially affect the provider’s ability to provide the service. We offer Notice of any incident affecting the service without undue delay, and in any case within 24 hours of our becoming aware of it; and prompt written notice of any development that might materially affect our ability to provide it. (c) Contingency plans and security measures Requires Business contingency plans that are implemented and tested, and appropriate ICT security measures, tools and policies. In place today A daily restore drill, run automatically, that restores the production ledger into a fresh database and checks every entry; the engine can also run in your own account. Restore drill Running it yourself . The database plan keeps no backups of its own, no penetration test has been done, and no certification is held. Security (d) Threat-led penetration testing Requires Participation and full cooperation in your threat-led penetration testing (Articles 26 and 27). We offer Full participation and cooperation, on terms agreed in advance so the test does not disturb other customers. (e) Monitoring, access, inspection and audit Requires Unrestricted rights of access, inspection and audit for you, a third party you appoint, and your competent authority, with the right to take copies; alternative assurance levels where other customers’ rights are affected; cooperation during inspections; and details of their scope and frequency. We offer Those rights, on reasonable notice to us, and without notice where your competent authority requires it. Much of the evidence is already public and checkable without asking us: the log, its heads, the anchors and the software bill of materials . (f) Exit, with a transition period Requires Exit strategies, including a mandatory adequate transition period during which the provider keeps providing the service, so you can move to another provider or in-house. We offer On any termination, including ours, the service continues at your request for a transition period of up to six months, on the same terms. In place today The documented exit plan. Continuity and exit plan Facts for your register of information Discuss a pilot’s terms ## Sealed security logs: evidence after a breach · RiskRouter URL: https://riskrouter.eu/security-logs Security logs evidence for after an incident After a breach, the logs are the first thing an intruder rewrites. When something goes wrong, everyone turns to the logs: your own investigators, the authority you report to, a cyber insurer assessing the claim, a customer who asks what happened to their data. An intruder knows this, and editing or deleting the lines that show what they did is often the first thing they do. A log you can still edit is your word about what happened, not evidence of it. logseal seals your logs as they are written, segment by segment, into the evidence log. Each segment is archived on your own storage, and only a salted fingerprint of its record reaches us. Later, anyone can check offline that every sealed segment is exactly what was written, and that none has gone missing. When it is tested The investigation after a breach; the notifications NIS2 expects within 24 hours, 72 hours and a month; a cyber insurer assessing the claim; a customer or a court asking what happened to their data. What you will be asked for What your logs said before and during the incident, your incident timeline, and what you reported and when (NIS2, Article 23; GDPR, Articles 33 and 34). What each article asks What goes wrong today The logs sit on the systems that were compromised, or on a collector the same administrators can edit. Whoever got in may have changed them, and nobody can prove they did not. What changes Logs are sealed segment by segment as they are written, on a collector the watched systems cannot reach. Afterwards anyone can check that every sealed segment is exactly what was written and that none is missing. The log lines never leave your storage. Try it in two minutes Choose a log file from one of your own servers. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. In production, logseal does this continuously, segment by segment; see How it runs . What it proves No sealed segment was changed Each segment’s SHA-256 is inside a record sent to the log as soon as the segment closes and the log can be reached. Change one line afterwards and the segment no longer gives the sealed digest. None went missing Segments are numbered and each record names the one before it, so a removed, inserted or renumbered segment shows. logseal verify lists exactly which. When, independently of you The log’s signed heads are timestamped by two independent authorities (not qualified under eIDAS) and anchored in Bitcoin, so a segment is shown to have existed no later than a time nobody can move. What you reported, and when Seal each early warning, notification and report as you send it, and the copy you hold is shown to be the one you sent, when you say you sent it. Checkable without us The open-source verifiers check a record’s proof offline, and logseal verify checks the archive against your own journal. Nobody has to trust you, or us. What it cannot do It does not detect anything It is not an intrusion detection system or a SIEM. It keeps the evidence of an incident; it does not spot one. It cannot say who did it It does not identify an attacker or decide who was responsible. Investigators reach those conclusions from the logs; the seal makes sure those logs can be trusted. It cannot protect lines written after a takeover Once an intruder controls the machine that writes a log, what that machine writes from then on can be forged at the source. What the seal protects is everything sealed before. Run logseal on a log collector the watched systems cannot reach, not on each of them. It does not make anyone compliant NIS2 and the GDPR ask for incident handling, reports and documentation. Sealing shows when a record existed; whether the record is enough is yours to judge. What each article asks How it runs logseal is part of the drop-in recorder kit (Apache-2.0, no dependencies). It reads lines on standard input, so it takes anything that can print a log: journalctl , tail -F , a syslog relay, an export of cloud audit logs. journalctl -f -o short-iso | riskrouter-logseal --source auth-server --archive /var/lib/logseal \ --lines 1000 --seconds 60 --format syslog riskrouter-logseal verify --archive /var/lib/logseal --journal evidence-journal.jsonl # auth-server: 412 of 412 sealed segments intact, 412 recorded in the log A segment Closes after a number of lines or seconds, whichever comes first. It is written once to the archive, synced, and never overwritten; only then is its record sealed. The record A security.log-segment record of the NIS2 and data-breach pack : the source, the segment’s number, its time span, its line count and its SHA-256. The record and its salt stay in your journal; the log lines never leave your storage. A restart Carries on the same numbered chain from the last archived segment. An outage Records wait in your journal and are sent when the log answers again; nothing is lost, and nothing is marked recorded before the log has answered. The same pack records each stage of an incident, and each report or notification you send. Download: logseal.mjs , with the recorder it uses, from the recorder kit . Recording needs a key; a sandbox key is free, with no form and no contract. Who it is for Organisations under NIS2, which must handle incidents and report significant ones within 24 hours, 72 hours and a month; every controller under the GDPR, which must document each personal-data breach; and anyone who will one day have to show a cyber insurer, an authority or a court what their logs said before an incident. Financial entities report ICT incidents under DORA, whose pack covers them. ## AI agents and assistants: connect over MCP · RiskRouter URL: https://riskrouter.eu/ai For AI agents, assistants and the people who build them An AI decision you cannot show later is an AI decision you cannot defend. When an AI system advises, scores or acts, someone will one day ask what it did and why: a customer, a supervisor, a court. RiskRouter gives AI systems the same evidence log it gives brokers: each record stays with you, only its salted fingerprint is sealed, and anyone can check it offline. This page is everything an AI agent, an assistant or its builder needs, in one place. Connect an assistant in one step The evidence log is an MCP server at https://api.riskrouter.eu/mcp (Streamable HTTP). Reading needs no key. Recording needs your key in an Authorization header; a sandbox key is free, with no form and no contract. Claude Code claude mcp add --transport http riskrouter https://api.riskrouter.eu/mcp \ --header "Authorization: Bearer " Cursor, and any client configured in JSON { "mcpServers": { "riskrouter": { "url": "https://api.riskrouter.eu/mcp", "headers": { "Authorization": "Bearer " } } } } Claude The evidence log is listed in Claude’s connector directory: RiskRouter evidence log . Add it from there to read the log and fetch proofs, with no key. Other clients that add a server by its address Add a custom connector with the address above. If the client cannot send an Authorization header, it can still read the log and fetch proofs; record from Claude Code, Cursor or the local server below. Use it from your assistant Every assistant below reaches the same address, https://api.riskrouter.eu/mcp , and gets the same tools. Reading the log and fetching proofs needs no key; recording needs your key as a bearer token. Menus change often, so each entry links the vendor’s own instructions. Claude (web, desktop and mobile) Add RiskRouter evidence log from the connector directory, or add a custom connector with the address above. Claude Code The command in Connect an assistant above. ChatGPT Add the address as a custom MCP connector, where your plan allows one ( OpenAI’s instructions ). If the connector cannot send a key, it can still read the log and fetch proofs. Gemini CLI gemini mcp add --transport http riskrouter https://api.riskrouter.eu/mcp , adding --header "Authorization: Bearer " only to record ( Gemini CLI’s instructions ). Grok, through the xAI API Give the address as a remote MCP tool in your request ( xAI’s instructions ). OpenAI API Give the address as a remote MCP server tool in a Responses API request ( OpenAI’s instructions ). Claude API The MCP connector of the Messages API ( Anthropic’s instructions ): "mcp_servers": [{ "type": "url", "url": "https://api.riskrouter.eu/mcp", "name": "riskrouter", "authorization_token": "" }], "tools": [{ "type": "mcp_toolset", "mcp_server_name": "riskrouter" }] // with the header anthropic-beta: mcp-client-2025-11-20 Whichever you use, the record stays with you: the assistant sends only a digest and a short kind, and there is no pricing tool. What an agent can do with it Tool What it does Key get_evidence_head The log’s current signed head: its size, root and signature No get_inclusion_proof The proof that one of your entries is in the log Yes get_consistency_proof The proof that the log only grew between two heads No get_checkpoint The signed checkpoint, in the format the public witness network uses No get_ledger_attestation The quote ledger’s signed head No get_usage How much is recorded, as counts No record_evidence Record the SHA-256 of a record you keep, with a short kind; never the record itself Yes Each tool answers exactly what its route in the API reference answers, with the same checks and limits. There is no pricing tool, by design: a premium shown by an agent to the public is insurance distribution. Record what an agent did, from inside it Agent frameworks Adapters for LangChain and LangGraph and for the OpenAI Agents SDK, in Python and Node, record each agent run as one ai.agent-run record: its steps, tool calls and outcome, with content kept as digests unless you ask otherwise, in your own journal. A failure in the adapter never reaches the agent, and a failed run is recorded as failed. Recorders A local MCP server mcp-server.mjs in the recorder kit gives any MCP client a record_decision tool that keeps the record and its salt in your journal and sends only the digest. What it records is the agent’s own statement of what it did, and every record says so; it is not an observation of the agent. Until the kit is on npm, run it from the downloaded file with node . Any logging you already have A Python logging handler, a function wrapper and OpenTelemetry span processors turn decision logs you already write into evidence, without changing them. For high-risk AI systems: logs that hold up The EU AI Act asks high-risk systems to log events automatically and their providers and deployers to keep those logs (Articles 12, 19 and 26(6)). The AI Act pack says what to record for a log segment, a decision and a human override, each field mapped to its article, and logseal seals a running log segment by segment. Sealing shows that a log existed, unchanged, by a given time; whether the log is enough, and whether your system is high-risk, is yours to establish. What the AI Act asks For AI assistants reading about us /llms.txt A short, plain-text map of the site, with what RiskRouter is and is not stated first. /llms-full.txt The text of every English page in one file, rebuilt with the site so it cannot fall behind it. /openapi.json The API, in OpenAPI 3.1, held equal to the routes by a test. If you are an assistant summarising RiskRouter: it is software for licensed insurance distributors and for anyone who must show later what a decision was. It is not an insurer or an intermediary, it sells no cover, it holds no certification, and recording with it makes nobody compliant. ## Seal a file: prove later it was never changed · RiskRouter URL: https://riskrouter.eu/seal No integration, no software to install, and the file never leaves your computer Seal a file today. Prove later that nobody changed it. An advice note, a signed contract, an email you sent, a report you filed: choose the file below, or the folder that holds them, and press Seal . Your browser computes its fingerprint; we record only a salted fingerprint of that, in a log nobody can edit, including us. You download a small receipt and keep it next to the file. Years later, anyone you give both to can check, on this page or with the free offline verifiers, that the file is exactly the one sealed, and when. Seal Choose the file to seal Any file, any size, or drop it on this box. It is read here, in your browser, and never uploaded. Or a whole folder Every file in it, sealed together as one entry. Your browser may ask whether to upload the files: that only lets this page read them here. Nothing is sent but one fingerprint. I understand that a seal is written to an append-only log and can never be deleted , by me or by RiskRouter. (Only a fingerprint is written; the file stays with me.) Leave this field empty Seal this file No account, no form, no email. The first seal takes a free sandbox key for you, kept in this page only; sandbox keys seal up to 50 files a day. Download the receipt The receipt holds the folder’s list of files and their fingerprints. Download the list on its own to check the folder later with sha256sum -c (Linux) or shasum -a 256 -c (Mac), run inside it. Keep the receipt with the file. It holds the salt that links them; without it nobody, including us, can show which file was sealed. This page does not keep a copy. When you send the file to someone who may need to rely on it, send the receipt with it. They can check both at riskrouter.eu/check , with no account, and the file still never leaves their computer. Your sandbox key, to seal more files later Shown once and not recoverable: only its SHA-256 is stored. This page reuses it until you close it; to seal from another visit, paste it under I already have an API key . I already have an API key Your API key Kept in this page only, never stored. Seals made with it are recorded against your account. Check Given a file and its receipt, this checks that the file is the one sealed, that the log recorded it, and that we signed the log’s state, against our published key. A folder is checked file by file: anything missing or changed fails. One file from a sealed folder can be checked against the folder’s receipt. Nothing is uploaded. The file Or the folder Its receipt (.json) Check What a seal proves, and what it does not It proves that this exact file existed no later than the moment its seal entered the log, and has not changed by a single byte since. The log’s signed state is time-stamped by two independent authorities (not qualified under eIDAS) and anchored in Bitcoin. It does not prove who wrote the file, that what it says is true, or that it satisfies any law. It shows the file is the one you had at the time; whether that file is enough is yours to judge. Checkable without us The receipt is a standard record bundle. The open-source verifiers check it offline: python3 riskrouter_verify.py receipt.json --key anchors/ . Verifiers and keys A folder is one entry in the log, however many files it holds (up to 10,000). The receipt lists each file’s path inside the folder and its SHA-256, in the format sha256sum writes, and that list is what is sealed. It proves those files were there, as they were; a file added later is not covered, and is said so when you check. .DS_Store , Thumbs.db and desktop.ini are left out: the system rewrites them when a folder is only opened. For many files Sealing by hand suits a few files a day. For every advice note your software writes, the same seal is one API call from inside it. Integrate ## Is this all of it? Proofs beyond one record · RiskRouter URL: https://riskrouter.eu/completeness For auditors, supervisors and the firms they examine Is this all of it? Proofs beyond one record A proof for one record shows that it was sealed and has not changed since. The next questions are harder. Is anything missing from the file you were shown? What did one field say, without handing over the rest? Were the records kept at all, checked on a sample nobody could choose? Did the other party seal the same record? Was the report sent in time? Each answer below is checked offline with the published verifiers, in JavaScript or Python, against our public keys, without asking us. None of them sends a record to the log: only salted fingerprints, as always. Is anything missing? Completeness chains : numbered records, and a signed list of every position the log holds. What did this field say? Selective disclosure : one field of a sealed record, the others still sealed. Were the records kept? Spot checks : a sample drawn from a Bitcoin block hash that did not exist when the plan was sealed. Did both sides agree? Co-sealing : the same record, signed and sealed by each party with its own key. Was it reported in time? Deadline proofs : a report’s seal time against the NIS2, GDPR and Cyber Resilience Act clocks. Missing records Is anything missing? Completeness chains A firm that shows ten intact records may have sealed eleven and kept back the awkward one. A time stamp cannot catch that, because it proves only what it is shown. So a firm puts its records in numbered chains, one per desk, product, model or system. Each record carries its chain, its position and the fingerprint of the record before it, and the log refuses a position out of turn, with nothing written. Anyone holding the chain’s tag asks the log what it holds in that chain: GET https://api.riskrouter.eu/api/v2/evidence/chain?chain_tag=<64 hex> The answer is signed with the same key as the log’s heads: how many positions the chain has as of a signed head, and when and where each was sealed. It names no firm and holds no record. The firm shows its records; the verifier checks every position from the first to the last, and names any position the log holds that nobody showed. What it does not prove: that a record was put in a chain at all. That is a policy for the firm, its auditor or its supervisor to set, and the recorders chain every record automatically once given a chain secret. On a log a firm runs itself, a chain is that firm’s own word unless independent witnesses sign its heads. The construction One field What did this field say? Selective disclosure A lender asked about one decision needs to show the score it used, not the applicant’s file. Sealed this way, a record is a set of salted commitments, one per field. Later the firm reveals one field’s value and that field’s salt; the verifier checks that value against the sealed commitment, and the record against the log. The other fields stay sealed. Each field’s salt comes from a secret the firm never discloses, so a hidden field with few possible values, such as approved or declined , cannot be found by trying them all. The names of the fields are visible; their values are not. node records.mjs seal-fields decision.json node records.mjs disclose decision.json --secret --salt --proof proof.json --fields score,outcome The construction Sampling Were the records kept? Spot checks Sending an auditor everything is costly and hands over personal data. A sample is cheaper, but one the firm picks proves nothing, and one the auditor picks can be argued with. So the sample comes from a number neither side can choose: the hash of a Bitcoin block mined after the auditor sealed a plan naming its height. The population is the chain’s signed length. The draw is fixed and anyone can repeat it; here it is with the hash of Bitcoin’s first block, for a chain of 1,000 records and a sample of five: node chains.mjs select --seed 000000000019d6689c085ae165831e934ff763ae46a2a6c172b3f1b60a8ce26f \ --population 1000 --sample 5 positions: 234, 243, 955, 993, 512 The firm shows those five records and no others, and the verifier checks each against the chain. One step stays with the auditor: confirming, on any Bitcoin node or block explorer, that the block at that height has that hash. The verifiers contact nothing, so they say so rather than look it up. The draw Two parties Did both sides agree? Co-sealing Disputes are usually between two parties who each kept their own version: broker and insurer, lender and bureau, platform and customer. A record only one side sealed is that side’s word. When both share the record and its salt, each seals the same fingerprint as an entry signed with its own registered key. The verifier checks both entries, that they carry the same fingerprint, and that the keys and the parties are different; given each party’s key as it published it, that each entry was signed with exactly that key. It shows that each party asserted the same record at the time it claimed. It does not show the record is true, or that a key was not stolen. The check Deadlines Was it reported in time? Deadline proofs NIS2, the GDPR and the Cyber Resilience Act set reporting clocks that start when a firm becomes aware of something. A sealed report record carries when the firm says it became aware, when it says it sent the report, and the fingerprint of the report as sent; the seal adds a time the firm did not choose. Report Deadline NIS2 early warning, incident notification 24 hours, 72 hours from becoming aware (Article 23(4)) NIS2 final report One month after the incident notification (Article 23(4)) GDPR personal-data breach notification 72 hours from becoming aware, where feasible (Article 33(1)) Cyber Resilience Act early warning, vulnerability notification 24 hours, 72 hours from becoming aware (Article 14(2)) Cyber Resilience Act final report 14 days after a corrective or mitigating measure is available (Article 14(2)) node deadline.mjs check early-warning.json notification.json --key keys/ The tool checks each report against both times, and fails a report that claims awareness after an earlier sealed record of the same incident. It does not decide whether a report was due, whether its content was enough, or whether a delay was justified. DORA’s deadlines are set by delegated acts and are not computed here. Verify it Check it yourself chains.mjs checks chains, completeness and spot-check bundles and co-seals; records.mjs checks disclosures; deadline.mjs checks reporting times. Standard-library Node, in the conformance kit with the code they import. riskrouter_verify.py checks all of them, written separately in standard-library Python. Both are held to the same published test vectors , and the conformance suite tests any other implementation against them. Its result is the implementer’s own statement, not ours. A software vendor can give its customers a verification page in its own brand : one HTML file that checks a receipt or a disclosure in the browser and is forbidden, by its own security policy, to send anything anywhere. None of this proves a record is true, or that you have met an obligation; it shows what was sealed and when, and whether what you were shown is all of it. Whether that is enough is for you, your auditor or your supervisor to judge. ## Press: facts, logo and contact · RiskRouter URL: https://riskrouter.eu/press For journalists, analysts and anyone writing about RiskRouter Press: what RiskRouter is, in facts you can check Everything on this page can be checked without asking us. The figures are computed from the repository when the site is built, the same way the rest of the site states them, and each links to where you can verify it yourself. Describing us In one line, and in one paragraph One line. RiskRouter is a log nobody can edit, in which a firm seals the fingerprint of a record it may one day have to defend, and anyone can later prove the record was never changed. One paragraph. Regulated firms are asked, sometimes years later, what they advised or decided and whether the record of it has been touched since: an insurance advice note, an investment suitability statement, a credit assessment, an AI system’s decision log, a security log before a breach. RiskRouter seals a salted fingerprint of each record in an append-only log whose heads are signed, time-stamped and anchored in Bitcoin; the record itself never leaves the firm. Anyone holding the record and its receipt can check it offline with open-source verifiers, without trusting RiskRouter. It is built for the software vendors that serve regulated finance, and for AI platforms. It is a validation build, run from Brussels. Either may be quoted as written. The name is one word with two capitals: RiskRouter. Figures Facts, from the repository First head anchored 13 September 2026, in Bitcoin block 967 830. How to check Time stamps 38 signed heads carry RFC 3161 time stamps from DigiCert timestamp service and Sectigo timestamp service. None is a qualified time stamp under eIDAS. What to record, by law Fourteen regime packs for thirteen EU laws, each field mapped to the article that asks for it. Regime packs Industries covered Nineteen pages, one per kind of record people are later asked to prove. Industries Independent checks Two verifiers written separately (JavaScript and Python, standard libraries only), published test vectors and a conformance suite. Specification Independent witnesses None yet. The log is served in the format the public witness network cosigns. Results published by others None yet. How much is recorded, and by whom, is counted live at /usage ; whether the service answers, measured every ten minutes from outside, is at /status . Every claim the site makes, and where to check it: /evidence . Not What RiskRouter is not Not an insurer or an insurance intermediary, and it sells no cover. The pricing engine on the site is a worked example for developers. Regulatory position Not certified by anyone, and its time stamps are not qualified under eIDAS. Why Not a company yet: it is run by one person, Anil Jangra, with no legal entity registered, and no commercial terms are in force. What that means if you depend on it Not a holder of anyone’s records: it receives salted fingerprints, never the records. Data protection Not a smart contract, a token or a guardrail: it enforces no limits on what a system or an AI agent does, and it is unrelated to the RiskRouter.sol contract used in ERC-8004 AI-trading-agent hackathon projects. Its only use of a blockchain is anchoring signed log heads in Bitcoin. It records evidence; it blocks nothing. RiskRouter names no customer, partner or user, because it has none it could name; please do not describe any organisation as one. Brand Name and logo The mark is three linked squares, the last one filled: a chain of records whose newest entry is sealed. The mark, SVG The mark, PNG, 512 512 The default share card, PNG, 1200 630 Colours: ink #18191b on paper #fbfaf7 , with one accent, #b3361b , for the filled square. The type is Source Serif 4 for headings and IBM Plex for text, both open fonts. Write Contact Write to anil@riskrouter.eu . You will be writing to Anil Jangra, who builds and runs RiskRouter, and the answer comes from the same person. Other ways to reach us ## A pilot that costs nothing and commits to nothing · RiskRouter URL: https://riskrouter.eu/pilot For software vendors considering a pilot A pilot that costs nothing and commits you to nothing When we ask a vendor whether it would sign a one-page pilot letter, this is what the letter says. It is a letter of intent, not a contract: apart from confidentiality and the right to end the pilot, it creates no obligation, no exclusivity and no fee. Read it here before you answer. What you do Integrate the API into a test or limited build of your product, using the published specification and, if you like, the integration kit . Run the conformance suite against your build and keep the result. Name a technical contact, and at the end say in writing what worked, what did not, and whether you intend to ship it. What we do Provide the API keys at no charge, a technical contact who answers within one working day, and the changes the pilot shows are needed, where they fit the product’s published rules. What it costs Nothing, for either side. No commercial terms are agreed in the letter. If you decide to ship, terms are discussed then, and nothing obliges either side to agree any. Data No personal data reaches us: your product sends a salted SHA-256 fingerprint of each record and a short record type; the record, its salt and the customer stay in your product. Entries in the log are permanent by design and cannot be deleted by anyone, including us; you tell the firms taking part. Data protection Phase 1, from signature Technical evaluation with sandbox keys only. No live records of real firms. Phase 2, live use Starts only when all three hold: your build passes the conformance suite; we hold the authorisation needed to carry on this activity in Belgium; and a specialist legal opinion confirms the activity needs no registration as an insurance intermediary. We tell you in writing when the last two are in hand. It lasts 90 days from then, longer if both sides agree in writing. Ending it Either side, at any time, by written notice, without cause and without cost. What was recorded stays recorded and verifiable, and you can export it at any time. Continuity Success For the number of firms taking part that the letter names, a record sealed from your product can be proven by that firm, against a signed head of the log, using only the public verifier and without contacting us; and you have said in writing whether you intend to ship. Names and results Neither side uses the other’s name in public without written agreement. Your conformance result is yours: you may publish it; we will not, unless you agree. Freedom No exclusivity. The specification is published so that anyone may implement it; nothing in the letter stops you doing so yourself, or using any other supplier. Law Belgian law; the courts of Brussels, for the few obligations the letter creates. On request The letter itself One page, in English, written first for the software brokers and insurers use; the same terms work for any software vendor serving regulated firms. A Dutch or French version is available. Have your own counsel read it: it is a template, and neither side should sign on the strength of a summary. Ask for the letter Start without it: a sandbox key, no form You do not need the letter to try anything. A sandbox key is free, needs no contract, and works today; the letter only records that a pilot is under way, and what each side has said it will do. ## Roadmap: what comes next, and what it waits on · RiskRouter URL: https://riskrouter.eu/roadmap For product, business and anyone deciding whether to build on RiskRouter Roadmap: what comes next, and what it waits on No dates we cannot keep. Each step below names the thing it waits on, so you can tell for yourself whether it has happened. What has shipped, and when, is in the changelog . Today Built and running today The evidence log: salted fingerprints in a Merkle tree, signed heads, RFC 3161 time stamps and Bitcoin anchors, checked offline by two independent verifiers. Verify Proofs beyond one record: completeness chains, selective disclosure, spot checks, co-sealing and deadline proofs. Is this all of it? What to record, by law: regime packs, each field mapped to its article. Regime packs Ways in: the API, integration kits in Python, PHP, C# and Java, drop-in recorders on PyPI and npm, agent adapters, an MCP endpoint, and sealing a file in the browser. Integrate Seal a file Running it yourself, with the same code, in your own account. Self-hosting An individual Internet-Draft on digest-only evidence logs, posted at the IETF on 8 October 2026; no working group has adopted it. draft-jangra-scitt-digest-evidence Next Next, and what each step waits on Step Waits on Where it stands Commercial terms, production keys for paying customers, a legal entity The authorisation needed to carry on this activity in Belgium Applied for. Until then the sandbox is free and no commercial terms are in force. Live pilots with software vendors That authorisation, a specialist legal opinion, and the vendor’s build passing the conformance suite Technical evaluation can start now. The pilot letter Independent witnesses co-signing the log The public witness network listing the log, then witnesses configuring it Requested. Witnesses shows the count from the files: none yet. Quote records in the Merkle log at volume Real load on the original ledger reaching its ceiling for a contracted firm Designed, not built: nothing is built for load that does not exist. Outreach beyond two segments The review of 4 November 2026, on counted results Software vendors serving regulated finance, and AI platforms, until then. Not now Decided against, for now Qualified time stamps Need a paid contract with a qualified authority. Taken up when a customer needs one; the tooling to request and check them is already in place. Why A qualified electronic ledger under eIDAS Needs a legal entity, an audited trust service and a paid conformity assessment renewed every two years. Revisited when the entity exists and a customer, a tender or a supervisor asks for one. Why Zero-knowledge proofs over many records Revisited when a supervisor, an auditor or a pilot firm names a rule it wants proven without seeing the records. Selective disclosure and completeness chains cover most of the need today. What exists Pages in French, Dutch and German Withdrawn on 5 October 2026: the site is in English for now. Bringing one back is a switch in the build, which then refuses any missing string. Never What RiskRouter will never build Claims handling, customer care, legal services or carrier management. We never touch your customer or own your carrier relationship: those are boundaries, and they are why you keep control. Lock-in: no SDK you are required to use, no format only we can read. The specification is open, and anyone may implement it. A sandbox behind a sales form, or a feature built only to fill a comparison table. A claim of a licence, a certification or a customer we do not have. Your say Tell us what is missing If a step here would decide whether you build on RiskRouter, say so: a concrete need from a vendor goes to the top of the list, inside the rules above. Contact ## Investment advice: suitability you can prove · RiskRouter URL: https://riskrouter.eu/for-investment-advice For investment firms, private banks and financial advisers When a client says the investment was unsuitable, show the assessment as it was on the day. Markets fall, and some clients then remember a different conversation. What settles it is the suitability assessment and the statement on suitability you gave before the transaction. What weakens it is any doubt that those documents are the ones you had at the time, rather than ones tidied up once the loss was known. Try it in two minutes Talk about a pilot When it is tested A complaint after a loss, taken to your national financial ombudsman; a supervisor’s inspection of your advice files; a claim against your professional liability cover; a court. What you will be asked for The information you obtained on the client’s knowledge, experience, financial situation and objectives (MiFID II, Article 25(2)), and the statement on suitability given before the transaction (Article 25(6)). Delegated Regulation (EU) 2017/565, Article 72(1), describes how records are to be kept: so that corrections, and the contents before them, can be easily ascertained, and so that records cannot otherwise be manipulated or altered. Where Article 16(7) applies, the recordings of telephone conversations and electronic communications, kept for five years, or up to seven where the supervisor asks. What goes wrong today The assessment lives in an advice tool, a CRM or a document store, and any of them can be edited by someone with the right access. Their audit trail is that system’s own word. A risk profile updated innocently after a loss looks exactly like one changed to win the complaint. What changes Each assessment and each statement is sealed the moment it is produced, by your advice software or by hand. A correction becomes a new sealed version, so the original stays provable beside it. In a dispute you hand over the document and its receipt, and the other side checks both themselves. The client’s data never reaches us: only a salted fingerprint does. A call recording stays in your archive; its fingerprint is sealed as the call ends, so the file you produce years later can be shown to be the one recorded. Sealing shows that a record existed unchanged from a given moment; whether your record-keeping meets Article 72 is yours to judge. The MiFID II record template lists the fields to keep and the article each one answers. Try it in two minutes Choose a suitability report or a risk-profile questionnaire of your own. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own advice software it is one API call per document. Integrate The case for your software vendor ## For automated and AI decisions: logs you can prove · RiskRouter URL: https://riskrouter.eu/for-ai-decisions For teams running credit, hiring, claims or other automated decisions, and AI agents When someone asks why the system decided, show what it did, unchanged. An applicant refused, a candidate screened out, a claim declined, an agent that took an action nobody expected. The question that follows is always the same: what exactly happened, which version of the system did it, and can you show the log was not edited once the question was asked? Try it in two minutes Connect an agent over MCP When it is tested A person contests an automated decision about them; a supervisor or market-surveillance authority asks for your logs; an internal review after an incident; a court. What you will be asked for For high-risk AI systems, the events logged automatically (AI Act, Article 12), kept by the provider (Article 19) and by the deployer for at least six months (Article 26(6)). For a decision based solely on automated processing, the person’s right to human intervention and to contest it (GDPR, Article 22(3)), which you can only honour if you can say what was decided and why. What goes wrong today Logs rotate, are overwritten, or sit in the same infrastructure the team can edit. Model versions move on. Months later nobody can say which version decided, and an honest log is no easier to believe than an edited one. What changes One sealed record per decision or agent run: the inputs’ fingerprint, the model version, the outcome and any human review, kept in your own systems with only a salted fingerprint sent to us. Drop-in recorders for your existing logs, LangChain and the OpenAI Agents SDK do it without changing your code’s logic. AI agents and assistants Sealing shows that a log existed unchanged from a given moment. Whether your system is high-risk, and whether its logs are enough, is yours to establish. What the AI Act asks GDPR Article 22 If you build an AI-governance or deployment platform Your customers already keep their inventory, assessments and logs in your platform. What they cannot show from inside it is that none of it was rewritten after a decision was contested, because your platform, like any system, can be edited by whoever runs it. What you add One call when a log segment, an assessment or a model card is finalised: its salted fingerprint goes to the evidence log, the content stays in your platform. Each customer can then prove, to a supervisor or a court, that the record is the one from the day. Why it is not a competitor We hold no models, no logs and no assessments, and we build no governance features. One open format, free verifiers, no lock-in. Our boundaries How to start A free sandbox key with no form, the recorders for Python and Node, or the hosted MCP endpoint . A first integrating platform talks directly to the person who builds the API. If you train or fine-tune models: seal each checkpoint without showing it The question after an incident is often not what the model decided but which model it was: the exact weights, and the evaluation they passed before release. Weights are the last thing a lab will hand over, and an internal registry can be rewritten by whoever runs it. What you seal When a checkpoint is saved or released, a small record naming it: the SHA-256 of each weights file, the training configuration’s fingerprint, and the evaluation report’s fingerprint with its headline scores, if you choose to include them. The record stays with you; only its salted fingerprint is sent. What you can show later That the weights you hand an auditor, a customer or a court are byte for byte the ones you sealed on that date, with that evaluation, without having shown them to anyone before, us included. Anyone can check it offline with the free verifiers. How The recorders or the release-sealing GitHub Action in a training pipeline, one entry per checkpoint; or /seal on a folder of files by hand. Large files are fingerprinted where they are; nothing is uploaded. And the data it learned from The ai.training-data record seals each training, validation or test data set by its manifest, with your records of where the data came from and how it was prepared (AI Act, Article 10(2)) and, for a general-purpose model, of your copyright check and published training summary (Article 53(1)(c) and (d)). The data never leaves you. A seal proves which files existed and when, not that the evaluation was sound or that a model is safe. Try it in two minutes Choose a decision log export or an agent run's output of your own. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. In production, the recorders seal every decision as your system writes its log. ## Industries: whatever you may have to defend · RiskRouter URL: https://riskrouter.eu/industries By industry and by function Whatever you may one day have to defend, seal it the day you make it. The records differ from one field to the next; the moment they are tested does not. Someone asks for what you had on the day, and the question is whether anyone could have changed it since. Each page below says when that happens in your field, what you will be asked for, and how to try it on a file of your own in two minutes. Financial services Insurance brokers and agents Why you recommended this policy to this client, years later: each demands-and-needs note sealed the moment it is written (IDD, Article 20). Insurers and MGAs Show what you sold, how each product was approved and what you reported, exactly as it was, checkable without trusting you or us. Investment firms and financial advisers The suitability assessment and the statement on suitability, exactly as they were before the transaction (MiFID II, Article 25). Lenders and credit platforms When a borrower says the loan should never have been granted, or refused, show the assessment exactly as it was made, checkable by anyone. Banks, payment and crypto firms (AML and KYC) Show that each customer check was done, when, and on what evidence, exactly as recorded at the time, without sharing customer data. Software vendors serving any of these One integration gives every customer a record they can show a supervisor, an auditor or a court. You keep the customer, the data and the interface. Any organisation Teams running AI or automated decisions What the system decided and which version decided it, from a log nobody could rewrite afterwards (AI Act, Articles 12, 19 and 26; GDPR, Article 22). HR teams screening candidates with software When a candidate asks how they were assessed, show the screening record and the human review exactly as they were, without sharing their data. Security and IT leads What your logs said before an incident, sealed segment by segment as they are written (NIS2, Article 23; GDPR, Article 33). Data protection officers Show that breaches were documented, assessments made and consents recorded when you say they were, without sending us any personal data. Finance teams issuing and receiving invoices Seal each invoice issued or received, so in a tax audit or a dispute you can show the copy you kept is the one that existed at the time. Beyond financial services Software makers Show exactly what you shipped, its bill of materials and when you learned of a vulnerability, from signed statements the artifact never leaves. Telecom operators and network providers Seal each configuration change and each log segment as it is written, so after an incident you can show what changed and when, from evidence anyone can check. Pharmaceutical, biotech and testing laboratories Seal each result and batch record as it is created, so data integrity rests on evidence anyone can check, not on your audit trail alone. Clinical trial and medical device software Seal each trial document, safety report and device technical file as it is made, so any later alteration shows. The record never leaves your systems. Companies publishing a sustainability report Seal each dataset and each version of the report, so the figures behind your sustainability statement can be shown as they were when reported. Researchers, designers and inventors Seal a document, design, dataset or code without publishing it, and prove later that you had it by a certain date, without trusting us. Contracts, minutes and board decisions Seal each contract version, set of minutes and board resolution, so you can show later exactly what was agreed or decided, and when. Contracting authorities Seal each step of a procurement procedure as it closes, so award decisions can be shown later exactly as they were made and justified. Not listed? Any record works the same way: the same call with a different kind , or a file sealed by hand on Seal a file . The record stays with you; only a salted fingerprint reaches us. The quickstart Tell us what you need to prove ## Insurers and MGAs: records you can prove · RiskRouter URL: https://riskrouter.eu/for-insurers For insurers, MGAs and their compliance teams When your sales or product decisions are questioned, show the record as it was on the day. A complaint about a policy sold directly, a supervisor reviewing how a product reached its target market, an ICT incident under review. Each turns on records your own systems hold, and on whether anyone could have changed them since. Try it in two minutes Talk about a pilot When it is tested A complaint escalated to the insurance ombudsman; a supervisory inspection of your distribution and product governance; a review of an ICT incident and of what you reported. What you will be asked for When you sell directly, the customer’s demands and needs and any advice given (IDD, Article 20); the approval process and target market of each product (IDD, Article 25); for an ICT incident, how it was handled and what was reported (DORA, Articles 17 and 19). What goes wrong today These records sit in policy administration, CRM and ticketing systems that administrators can edit. Their audit trails are those systems’ own word, and an outside reviewer has to take it. What changes Each advice note, product-approval decision, product version and incident report is sealed as it is made, from your systems through the API or by hand on this site. A later correction is a new sealed version beside the original. A broker and an insurer can each seal the same record under their own keys, so a dispute starts from a version both put their names to. Nothing about your customers reaches us: only a salted fingerprint. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. What the IDD asks What DORA asks Co-sealing Try it in two minutes Choose a real file of your own: a product approval record, an advice note or an incident report. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Lenders: credit decisions you can prove · RiskRouter URL: https://riskrouter.eu/for-lenders For banks, consumer and mortgage lenders, and credit platforms When a borrower contests a credit decision, show the assessment exactly as it was made. A borrower in arrears says the loan should never have been granted. An applicant refused by a scoring model wants to know why. Either way the answer is the creditworthiness assessment on the day, and the doubt is whether it still reads as it did. Try it in two minutes Talk about a pilot When it is tested A complaint or court case about irresponsible lending; an applicant contesting an automated refusal; a supervisor reviewing your credit files. What you will be asked for The creditworthiness assessment and the information it rested on (Consumer Credit Directive (EU) 2023/2225, Article 18(1); for mortgages, Directive 2014/17/EU, Article 18); where the assessment involves automated processing, the consumer’s right to human intervention, an explanation and a review of the decision (Article 18(8)); for a decision based solely on automated processing, what was decided and the human review the person may ask for (GDPR, Article 22). What goes wrong today Scores are recomputed when models change, inputs are refreshed, and the file shows today’s data rather than what the decision saw. An assessment rebuilt after the fact proves nothing. What changes Each assessment is sealed when the decision is made: a fingerprint of the inputs, the model or policy version, the outcome and any human review. The file stays in your systems; only a salted fingerprint reaches us. Asked about one decision, you can disclose the score it used and keep the rest of the file sealed. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. What the Consumer Credit Directive asks GDPR Article 22 One field, not the whole file Try it in two minutes Choose a real file of your own: a credit assessment export or a decision letter. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## AML and KYC: due diligence you can prove · RiskRouter URL: https://riskrouter.eu/for-aml-kyc For banks, payment and e-money institutions, crypto-asset service providers and other obliged entities When the supervisor asks whether you checked the customer, show the check as it was done. An AML inspection, a lookback after a suspicious case, a request from the financial intelligence unit. The question is not only whether a file exists, but whether it existed before anyone knew it would be looked at. Try it in two minutes Talk about a pilot When it is tested An anti-money-laundering inspection; a lookback exercise; a request from the financial intelligence unit or a court. What you will be asked for The customer due diligence documents and information, and the records of transactions, kept for five years after the business relationship ends (Directive (EU) 2015/849, Article 40); from 10 July 2027, the same under Regulation (EU) 2024/1624, Article 77, which also asks you to check customers against targeted financial sanctions and to check again regularly (Articles 20(1)(d) and 26); for ICT incidents affecting these systems, how they were handled and reported (DORA, Articles 17 and 19). What goes wrong today KYC files are updated, re-scanned and remediated in batches. Afterwards nobody can show which documents were on file when the account was opened, and honest remediation looks like back-dating. What changes Each completed check is sealed at the moment it completes: a fingerprint of the documents seen, the screening result and the reviewer. Seal the version of each sanctions list you screen against, and a result can be read against the list as it stood that day. Remediation adds new sealed versions, and the original stays provable beside them. Customer data never reaches us. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. What the Anti-Money Laundering Regulation asks What DORA asks Try it in two minutes Choose a real file of your own: a completed due-diligence checklist or a screening report. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Hiring with software: screening you can prove · RiskRouter URL: https://riskrouter.eu/for-hiring For HR teams, recruiters and the software they use to screen candidates When a candidate asks how they were assessed, show what the screening did, unchanged. A rejected candidate asks for an explanation, alleges discrimination, or exercises their data rights. AI used to recruit or select people is high-risk under the AI Act, and the logs of what it did are the first thing anyone asks for. Try it in two minutes Talk about a pilot When it is tested A candidate’s complaint or discrimination claim; a data-protection request; a labour inspectorate or market-surveillance authority asking for logs. What you will be asked for For AI used to recruit or select people, a high-risk use (AI Act, Annex III, point 4): the logs it generates, kept by the deployer for at least six months (Article 26(6)); for a decision based solely on automated processing, the candidate’s right to human intervention and to contest it (GDPR, Article 22). What goes wrong today Screening tools keep their logs inside the vendor’s platform, criteria change between rounds, and months later nobody can show which version ranked whom, and why. What changes Each screening decision and each human review is sealed as it is made, with the version of the tool and the criteria used. Seal the screening criteria themselves when they are set or changed, and a decision can be read against the criteria in force that day. The candidate’s data stays in your systems; only a salted fingerprint reaches us. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. What the AI Act asks Screening criteria, field by field Automated and AI decisions Try it in two minutes Choose a real file of your own: a screening report or an interview evaluation. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Data protection: accountability you can prove · RiskRouter URL: https://riskrouter.eu/for-data-protection For data protection officers and privacy teams When the authority asks what you knew and when, show your records as they were written. After a breach, the first questions are when you found out, what you decided and when you notified. Under the GDPR you must be able to demonstrate compliance, and a record that could have been written last week demonstrates little. Try it in two minutes Talk about a pilot When it is tested A data protection authority’s investigation after a breach or a complaint; a data subject’s claim; an audit of your records of processing. What you will be asked for That you can demonstrate compliance (GDPR, Article 5(2)); that a person consented, where you rely on consent, and that withdrawing was as easy as consenting (Article 7(1) and (3)); erasure without undue delay where a ground applies (Article 17(1)); your records of processing activities (Article 30); every personal-data breach documented, with its facts, effects and the remedial action taken (Article 33(5)); data protection impact assessments (Article 35). What goes wrong today Breach registers and impact assessments live in spreadsheets and document tools that anyone with access can quietly update, and their dates are whatever the file says. What changes Each consent and withdrawal, breach entry and assessment is sealed when written, and each erasure when carried out, as a record that names the systems and the ground and holds none of the erased data. Corrections become new sealed versions beside the original. Nothing personal is sent: only a salted fingerprint. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. GDPR consent and erasure GDPR breach articles, with NIS2 Was the breach reported in time? Try it in two minutes Choose a real file of your own: a breach register entry or an impact assessment. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Invoices: show the copy you kept is the original · RiskRouter URL: https://riskrouter.eu/for-invoices For finance teams, and the software that issues and receives invoices When a tax audit or a customer questions an invoice, show the copy you kept is the one you issued. Years after it was issued, an invoice is checked by a tax inspector, contested by a customer, or set beside a supplier’s copy that reads differently. The question is whether the copy in your archive is the one that existed at the time. Try it in two minutes Talk about a pilot When it is tested A VAT audit; a customer or supplier disputing an amount or a date; an investigation into altered or duplicated invoices. What you will be asked for That the authenticity of origin, the integrity of content and the legibility of each invoice are ensured from issue to the end of its storage period, for example by business controls that create a reliable audit trail between an invoice and a supply (VAT Directive 2006/112/EC, Article 233(1)); that copies of invoices issued and received are stored (Article 244), for a period each Member State sets (Article 247(1)). What goes wrong today Archives are migrated, PDFs are regenerated from the accounting system, and e-invoicing formats change. A regenerated invoice may be correct and still not be the one issued, and nobody can show which. What changes Each invoice is sealed when it is issued or received: a fingerprint of the file exactly as sent, with its number and date. The invoice stays in your archive; only a salted fingerprint reaches us. Numbered in a chain, a series can be shown to be complete: an invoice sealed and later left out shows as a gap. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. A seal is not an electronic signature or an e-invoicing format, and replaces neither. What the VAT Directive asks Showing nothing is missing Try it in two minutes Choose a real file of your own: an invoice you issued, as a PDF or an XML e-invoice. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Software releases: what you shipped, provably · RiskRouter URL: https://riskrouter.eu/for-software-releases For software makers and anyone placing products with digital elements on the EU market When a vulnerability is exploited, show what you shipped and when you knew, unchanged. A customer is breached through your product. The questions are which build they ran, what was in it, and when you learned of the flaw. The Cyber Resilience Act turns those questions into obligations. Try it in two minutes Talk about a pilot When it is tested An incident in a customer’s environment; a market-surveillance authority’s request; a dispute over when a vulnerability was known. What you will be asked for Documented vulnerabilities and components, including a software bill of materials (Cyber Resilience Act, Regulation (EU) 2024/2847, Article 13 and Annex I); technical documentation drawn up before the product is placed on the market and kept up to date (Article 31); for an actively exploited vulnerability, since 11 September 2026, an early warning within 24 hours, a notification within 72 hours and a final report no later than 14 days after a corrective or mitigating measure is available (Article 14(2)). What goes wrong today Build artifacts, SBOMs and advisories are kept in repositories and ticket systems the team controls, and can be regenerated or edited after the fact. What changes Each release, its SBOM and each advisory is registered as a signed statement through the IETF SCITT interface. The artifact never reaches us: only its hash, under your own signature, inside the log. Receipts are standard COSE receipts. Each vulnerability report is sealed as sent, and a deadline check compares its seal with the 24-hour, 72-hour and 14-day clocks. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. What the Cyber Resilience Act asks SCITT in the specification Deadline proofs Try it in two minutes Choose a real file of your own: a release artifact or its SBOM, or use ours: the real SBOM of the Workers behind this API . It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Telecoms: network changes you can prove unaltered · RiskRouter URL: https://riskrouter.eu/for-telecoms For providers of public electronic communications networks and services After an incident, show which network change came first, exactly as it was made. When a network goes down or is breached, the first questions are what changed, when, and whether the record of it is the one from the day. The change log usually lives on the very systems that were changed. Try it in two minutes Talk about a pilot When it is tested A significant-incident notification and its final report; a supervisory inspection; a dispute with a supplier over a change; your own post-incident review. What you will be asked for NIS2 applies, whatever their size, to providers of public electronic communications networks or of publicly available electronic communications services (Directive (EU) 2022/2555, Article 2(2)(a)(i)); those that qualify as medium-sized enterprises are essential entities (Article 3(1)(c)). Their measures must include incident handling, security in the maintenance of network and information systems, and access control policies (Article 21(2)(b), (e) and (i)), and significant incidents must be notified (Article 23). What goes wrong today Configuration archives and change logs sit on management systems that an administrator, or anyone holding their access, can edit; an export is a file anyone can change; and a change log is still the network’s own word. What changes Each configuration change, as committed, and each segment of the routers’ and management systems’ logs is sealed as it is written, with logseal and the nis2 pack. Only a salted fingerprint leaves your network; the configurations and log lines stay in your own archive. Numbered in a chain, the changes to one device can be shown complete: one sealed and later removed shows as a gap. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. A seal is evidence of what was written and when, never detection: it cannot say a change was authorised, identify an attacker, or protect what is written after a machine is taken over. Sealed security logs Try it in two minutes Choose a real file of your own: a router configuration export. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Laboratories: results you can prove unaltered · RiskRouter URL: https://riskrouter.eu/for-laboratories For pharmaceutical, biotech, clinical and testing laboratories When an inspector questions your data, show each result exactly as it was recorded. Data integrity findings are among the most common in inspections. The question is always the same: is this result the one the instrument produced, and was nothing changed or deleted since? Try it in two minutes Talk about a pilot When it is tested A GMP inspection; the review of a regulatory submission; a batch release dispute; an internal data-integrity investigation. What you will be asked for For computerised systems used in GMP, consideration, based on a risk assessment, of a system-generated audit trail recording all GMP-relevant changes and deletions (EU GMP, Annex 11, section 9). What goes wrong today Audit trails live inside the same systems and are configured by their administrators; exports are files anyone can edit; and an audit trail is still the system’s own word. What changes Each result, batch record or export is sealed when it is created. It does not replace your audit trail; it makes it checkable from outside, because any later change to a sealed file shows. Numbered in a chain, the results of a study or a batch can be shown complete: one sealed and later left out shows as a gap. A contract laboratory and its sponsor can each seal the same report under their own keys. The data stays in your systems. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. A seal is evidence beside your validated systems, not a validated system; whether and how it fits your quality system is your quality unit’s decision. Is this all of it? Try it in two minutes Choose a real file of your own: a result export or a batch record. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Clinical trials and devices: records you can prove · RiskRouter URL: https://riskrouter.eu/for-life-sciences For sponsors, investigators, device manufacturers and the software they use When an inspector asks for the trial master file or the technical documentation, show each document exactly as it stood. A trial master file is kept for at least 25 years after the trial ends; a device’s technical documentation for at least ten after the last unit is placed on the market. Over that time documents are revised, systems migrated and archives moved, and the question is always whether the version you produce is the one that existed on the day. Try it in two minutes Talk about a pilot When it is tested An inspection of the trial master file; a question about a safety report years after the trial; a notified body or competent authority asking which version of the technical documentation applied on a given date; an incident report that is questioned. What you will be asked for For a clinical trial, a trial master file that at all times holds the essential documents, archived for at least 25 years after the end of the trial on media that keep it complete and legible, with any alteration traceable (Regulation (EU) No 536/2014, Articles 57 and 58), and records of the adverse events reported (Article 41). For a device, technical documentation kept up to date and available for at least 10 years after the last device is placed on the market, 15 for implantable devices (Regulation (EU) 2017/745, Article 10(4) and (8); for in vitro diagnostics, Regulation (EU) 2017/746, Article 10), and serious incidents reported within 15 days of becoming aware of them (Article 87(3)). AI software that is itself such a device, or a safety component of one, and needs a third-party conformity assessment is a high-risk AI system (AI Act, Article 6(1)), with the logs that follow (Articles 12 and 19). What goes wrong today Document systems keep an audit trail, but it lives in the same system and is configured by its administrators; archives are exported, migrated to new media and handed to new owners; and a file produced years later is still the system’s own word. What changes Each essential document, safety report or version of the technical documentation is sealed when it is filed or released, and a new version names the one it replaces, so an alteration shows as a new record rather than an overwrite. When a file is archived, its whole folder is sealed by one file list, so a file later left out or changed shows. A sponsor and its contract research organisation can each seal the same document under their own keys. The documents, and anything about a subject or a patient, stay in your systems; only a salted fingerprint reaches us. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. A seal is evidence beside your validated systems, not a validated system; whether and how it fits your quality system is your quality unit’s decision. What the Clinical Trials Regulation asks The MDR and IVDR GMP laboratories Try it in two minutes Choose a real file of your own: a trial master file folder or a technical documentation export. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Sustainability reporting: figures you can prove · RiskRouter URL: https://riskrouter.eu/for-sustainability-reporting For companies reporting under the CSRD, and those who assure them When your sustainability figures are questioned, show the data as it stood when you reported. Sustainability statements are now published and assured alongside the financial statements. When a figure is challenged, by the assurance provider, an investor or an NGO, the question is what the data said when you reported it. Try it in two minutes Talk about a pilot When it is tested The assurance engagement; a challenge from an investor, an NGO or a journalist; a restatement the following year. What you will be asked for The sustainability information reported under the Corporate Sustainability Reporting Directive (Directive (EU) 2022/2464), and the data and evidence behind it that the assurance provider examines. What goes wrong today Data is collected in spreadsheets and ESG tools that keep changing as suppliers send corrections; months later nobody can tell which version a published figure came from. What changes Each dataset, calculation and report version is sealed when it is finalised. A correction is a new sealed version, so a restatement shows as a restatement and not as a silent change. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. Try it in two minutes Choose a real file of your own: an emissions dataset or a draft of the report. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Research and inventions: prove what you had, when · RiskRouter URL: https://riskrouter.eu/for-research-and-ip For researchers, designers, inventors and R&D teams When someone claims they had it first, show that you had it, on a date nobody can move. A design copied, a dataset disputed, a former partner claiming an idea. The quickest way to settle it is to show the exact file was in your hands by a certain date, without having had to publish it. Try it in two minutes Talk about a pilot When it is tested A dispute with a former partner or employee; a design or copyright dispute; a trade-secret case; a question of who contributed what in a collaboration. What you will be asked for Evidence that a specific document, design, dataset or piece of code existed, unchanged, by a given date, and that it was in your possession then. What goes wrong today File dates are trivially changed, emails can be edited and forwarded, and publishing something to prove its date gives it away. What changes Seal the file’s fingerprint and keep the file private. The date rests on our signed log, two independent time-stamps and a Bitcoin anchor, not on your word. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. A seal is not a patent or a registered design and gives no exclusive right; it is evidence that a file existed unchanged by a date. Try it in two minutes Choose a real file of your own: a notebook page, a design file or a dataset. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Contracts and board decisions you can prove · RiskRouter URL: https://riskrouter.eu/for-contracts-and-governance For legal, company secretarial and finance teams When someone disputes what was agreed or decided, show the version as it was. Which version of the contract was signed? What did the board actually resolve, and was it before or after the announcement? Disputes about agreements and decisions come down to versions and dates. Try it in two minutes Talk about a pilot When it is tested A contract dispute; a shareholder challenging a board decision; an auditor’s or an investigator’s review; disclosure in litigation. What you will be asked for The exact text agreed or decided, and its date: the signed contract and its annexes, minutes and resolutions, the version of a policy in force at a given time. What goes wrong today Versions multiply across email and document tools, and a file called final_v7_signed.pdf proves nothing about when it was made or whether it was the final one. What changes Seal each version when it is agreed, approved or signed. Anyone can later check that a document is the sealed version, and when it was sealed, without seeing any of your other documents. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. A seal is not an electronic signature and does not replace one; it proves a file existed unchanged from a moment. Try it in two minutes Choose a real file of your own: a signed contract, a set of minutes or a resolution. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Public procurement: award decisions you can prove · RiskRouter URL: https://riskrouter.eu/for-public-procurement For contracting authorities and public procurement teams When a bidder challenges an award, show each step of the procedure as it was. A losing bidder challenges the award and asks how the bids were scored. What settles it is the record of the procedure, made as it went, and certainty that the scoring was not adjusted once the challenge arrived. Try it in two minutes Talk about a pilot When it is tested A bidder’s challenge or review procedure; an audit by a court of auditors; an anti-fraud investigation. What you will be asked for The documentation of the progress of each procurement procedure, sufficient to justify decisions taken at all its stages, kept for at least three years from the date of award, and the individual report on each contract awarded (Directive 2014/24/EU, Article 84). What goes wrong today Evaluation grids and scoring notes live in spreadsheets and e-procurement exports that can be edited before they are produced. What changes Seal each tender as it is received, and each document as its step closes: the opening record, each evaluator’s scores, the evaluation report, the award decision. An evaluation sealed before prices are opened shows it was fixed first. Anyone can later check that none was changed. Sealing shows that a record existed unchanged from a given moment; whether your records are enough is yours to judge. What the procurement directive asks Try it in two minutes Choose a real file of your own: an evaluation grid or an evaluation report. It is read in your browser and never uploaded. Seal it on Seal a file . A free sandbox key is issued on the page; only a salted fingerprint reaches the log. Make a copy, change one character in it, and check both against the receipt at Check . The copy fails; the original passes. From your own systems it is one API call per record. Integrate Other industries ## Witnesses of the evidence log · RiskRouter URL: https://riskrouter.eu/witnesses Witnesses checked by people who are not us Anyone can witness a log A signed head proves that a log's operator published it. It cannot prove that the operator never went back and rewrote an earlier entry. A witness can: each time it runs, it takes the current head, checks a consistency proof against the head it saved last time, and co-signs only if the log grew without changing anything that came before. A witness is worth something because it is not the operator, so it runs on the witness's own account, with a key only the witness holds. What a witness does One round, every hour Checks the head Fetches the log's signed head and checks the signature with a key it pinned the first time it saw the log. Checks the past If the log has grown, asks for a consistency proof and checks it against the root it saved itself, never one the log supplies. Co-signs Only if both hold, signs riskrouter-evidence-cosign|v2|witness_id|tree_size|root_hash|cosigned_at with its own key, the format in the specification , checked by both verifiers. Raises an alarm If the log contradicts a head it saw (a different root for the same size, a smaller tree, a proof that fails), it keeps both signed heads, publishes them, and stops co-signing that log. The alarm is never overwritten: two heads signed by the same operator that cannot both be true are evidence the operator cannot disown. Cannot be driven Its web side only serves what it has stored: its public key, its latest cosignature of each log, and any alarm. No request makes it call a log; only its own schedule does. A cosignature proves that someone other than the operator saw that head and checked it against every head they saw before. It says nothing about the records behind the head, and a witness that stops running proves nothing about the time after. The registry Every log, and everyone who witnesses one A witness can follow the registry and witness every log it lists. The registry is a convenience, not a source of trust: a witness pins each log's keys on first sight, and if the registry later drops or changes a pinned key, the witness raises an alarm instead of taking the new one. A registry may not point a witness at a private address. Logs Log Operator API Head-signing keys riskrouter RiskRouter https://api.riskrouter.eu dd4b4b394c95586a ( published ) Witnesses No witnesses yet. Nobody outside RiskRouter witnesses a log in this registry today, and until someone does, a rewritten history would be caught only by a holder who kept an older head. This page lists a witness only once they publish their key somewhere under their own control. The public witness network Our log also speaks the C2SP formats the public witness network uses: a checkpoint at /tlog/evidence-v2/checkpoint and hash tiles beside it ( specification ). Its key is api.riskrouter.eu/tlog/evidence-v2+05e20636+AZsv9x6CMqcAlnY+GmAtkFthrxjtaGKnVpQEH9ZAkL7J . We post each new checkpoint to 6 witnesses of that network; 6 have cosigned, and 20 cosigned checkpoints are kept in anchors/checkpoint/ . Rendered from registry/logs.json , witnesses/ and anchors/ when this page was built. The same data, as JSON: /registry.json . Become a witness Three ways to run one, each on your own account The witness is published in the public verifier repository , in witness/ , beside the verifiers. Every way starts by making a key; the private half stays with you. node witness/node.mjs init --out ./my-witness --name "Your organisation" # ./my-witness/private-key.b64 secret: goes into a secret store, never into a repository # ./my-witness/public-key.json publish it on your own site, and send it to us to be listed A Cloudflare Worker Copy witness/wrangler.toml.example , create a KV namespace, add the key as a secret and deploy. An hourly Cron Trigger witnesses every log in the registry; the Worker serves its cosignatures at /cosignatures//latest.json . The free plan is enough. A fork on GitHub Fork the verifier repository, copy witness/github-workflow.yml to .github/workflows/ , add the key as the secret WITNESS_PRIVATE_KEY . Every hour it witnesses and commits its cosignatures to witness-state/ in your fork, where anyone can read them. If it ever raises an alarm, the run fails after committing the evidence. Your own server node witness/node.mjs run --state ./state --every 3600 --port 8080 : the same witness, storing its state in a directory and serving it read-only. Node 20 or later, no dependencies. To be listed, send us your public-key.json , the https address where you publish the same key, and, if you serve your cosignatures, the address of your witness. We add a file to witnesses/ ; from then on a daily job collects your latest cosignature of our log, keeps it only if it verifies with your key and ours and is consistent with our current head, and counts it here. We list nobody who has not published their key themselves. Cross-witnessing Every instance can witness every other An organisation that runs the engine itself can switch on the witness in the same stack: docker compose --profile witness up -d . It follows the registry, so it witnesses our log and every other listed log, and it can also witness its own. A self-hosted log whose operator publishes its keys can be listed too, and is then witnessed by everyone who follows the registry. Each new participant makes every log in the registry harder to rewrite unnoticed, including ours. There is no fee, no contract and no account with us in any of this. A witness owes us nothing and we cannot switch one off. ## How old the log is, day by day URL: https://riskrouter.eu/age Evidence the age of the record How old the log is, day by day Anyone could rebuild this software. Nobody can give a new log a head that a time-stamping authority saw last month, or put one in a Bitcoin block that was mined last week. This page lists every head of ours that carries that kind of evidence, read from the files in anchors/ when the site was built, so each row can be checked without us. The evidence log Each row is a head we signed: the root of every entry recorded up to then. A time-stamping authority states when it saw the head; the newest head of each day also goes into Bitcoin. Newest first; 20 heads, the full roots in age.json . Head signed Entries Root Independent time (RFC 3161) Bitcoin 14 November 2026 07:33 UTC 798 6df054c5fb3e5463 DigiCert timestamp service, 14 November 2026 15:09 UTC Sectigo timestamp service, 14 November 2026 15:09 UTC block 975 975 7 November 2026 07:03 UTC 750 4c52bbc992e0fdbd DigiCert timestamp service, 7 November 2026 08:51 UTC Sectigo timestamp service, 7 November 2026 08:51 UTC block 974 942 2 November 2026 07:33 UTC 702 bb7b3785b014e7b3 DigiCert timestamp service, 2 November 2026 16:06 UTC Sectigo timestamp service, 2 November 2026 16:06 UTC block 974 281 1 November 2026 07:20 UTC 654 5c0ee0c4f9322129 DigiCert timestamp service, 1 November 2026 13:31 UTC Sectigo timestamp service, 1 November 2026 13:31 UTC block 974 136 31 October 2026 06:53 UTC 606 5ddae0d7597d4695 DigiCert timestamp service, 31 October 2026 12:41 UTC Sectigo timestamp service, 31 October 2026 12:41 UTC block 973 981 30 October 2026 06:59 UTC 558 7016b2a96833595c DigiCert timestamp service, 30 October 2026 13:03 UTC Sectigo timestamp service, 30 October 2026 13:03 UTC block 973 844 29 October 2026 06:55 UTC 510 d08c2a9dca7bb491 DigiCert timestamp service, 29 October 2026 13:10 UTC Sectigo timestamp service, 29 October 2026 13:10 UTC block 973 698 28 October 2026 06:47 UTC 462 84ed625a4e14e977 DigiCert timestamp service, 28 October 2026 08:39 UTC Sectigo timestamp service, 28 October 2026 08:39 UTC block 973 564 24 October 2026 05:39 UTC 414 8bf9ecbd5e1f2f7b DigiCert timestamp service, 24 October 2026 06:01 UTC Sectigo timestamp service, 24 October 2026 06:01 UTC block 972 923 23 October 2026 05:55 UTC 367 53765c9d51dc4f58 DigiCert timestamp service, 23 October 2026 07:43 UTC Sectigo timestamp service, 23 October 2026 07:43 UTC block 972 778 22 October 2026 05:57 UTC 326 7d4ff2371d9274f5 DigiCert timestamp service, 22 October 2026 07:45 UTC Sectigo timestamp service, 22 October 2026 07:45 UTC block 972 621 21 October 2026 05:54 UTC 285 b042237854d9d59b DigiCert timestamp service, 21 October 2026 07:49 UTC Sectigo timestamp service, 21 October 2026 07:49 UTC block 972 482 20 October 2026 08:52 UTC 244 036a815a6b13207f DigiCert timestamp service, 20 October 2026 13:40 UTC Sectigo timestamp service, 20 October 2026 13:40 UTC not anchored 20 October 2026 05:52 UTC 243 d8c12039a6db19c3 DigiCert timestamp service, 20 October 2026 07:45 UTC Sectigo timestamp service, 20 October 2026 07:45 UTC block 972 330 19 October 2026 05:58 UTC 202 7e304c5177590a41 DigiCert timestamp service, 19 October 2026 10:57 UTC Sectigo timestamp service, 19 October 2026 10:57 UTC block 972 172 18 October 2026 05:42 UTC 161 2cd2b0c0a5c9c099 DigiCert timestamp service, 18 October 2026 06:04 UTC Sectigo timestamp service, 18 October 2026 06:04 UTC block 972 079 17 October 2026 05:29 UTC 120 2288b99df5d2f2ac DigiCert timestamp service, 17 October 2026 05:42 UTC Sectigo timestamp service, 17 October 2026 05:42 UTC not anchored 16 October 2026 10:19 UTC 79 7a1ddbffb646b09b DigiCert timestamp service, 16 October 2026 12:35 UTC Sectigo timestamp service, 16 October 2026 12:35 UTC not anchored 16 October 2026 00:14 UTC 2 df429daa750fdfc2 DigiCert timestamp service, 16 October 2026 01:30 UTC Sectigo timestamp service, 16 October 2026 01:30 UTC not anchored 27 September 2026 20:12 UTC 0 e3b0c44298fc1c14 DigiCert timestamp service, 28 September 2026 09:46 UTC Sectigo timestamp service, 28 September 2026 09:46 UTC not anchored The quote ledger The worked example’s v1 ledger, anchored when its head moves. Anchored Entries Head Bitcoin 14 November 2026 64 753539a41344f4b2 block 975 811 7 November 2026 63 9abac6885eba8e0d block 974 942 2 November 2026 62 6b57747a821247d4 block 974 281 1 November 2026 61 1b2f14830d416463 block 974 136 31 October 2026 60 1a92d11affd46e74 block 973 981 30 October 2026 59 ebdf236eab6bb688 block 973 844 29 October 2026 58 b121f4ab60813ead block 973 698 28 October 2026 57 4600c824eca0a08d block 973 564 24 October 2026 56 d88ac8c4f7867f9a block 972 923 23 October 2026 55 3671b3c8e1c3f707 block 972 778 22 October 2026 54 70989284fccc01d2 block 972 621 21 October 2026 53 50cf8ba10bab5128 block 972 482 20 October 2026 52 c1731f4f74c396ed block 972 330 19 October 2026 51 5b7e91fb65c98c34 block 972 172 18 October 2026 50 637a922a44e29990 block 972 079 13 September 2026 46 f95974f59ca3a6dd block 967 830 Check a row yourself Every file named here is in anchors/ of the public verifier repository , copied there by the daily anchor run with the tools that check it. The Bitcoin anchor With the standard OpenTimestamps client, which shares no code with us: ots verify anchors/ots/evidence-head-N.payload.opentimestamps-a.ots anchors/ots/evidence-head-N.payload . The payload file is the head’s frozen string, riskrouter-evidence-head|v2|tree_size|root_hash|timestamp , and nothing else. Offline, without a Bitcoin node: node tools/head-anchor.mjs verify anchors/ots/*.ots or python3 tools/python/riskrouter_verify.py anchors/ots/evidence-head-N.payload.opentimestamps-a.ots . The time-stamp tokens node tools/tsa.mjs verify anchors/tsa/evidence-head-N.digicert.json --key anchors/ , or the same file with python3 tools/python/riskrouter_verify.py . Each token is over the SHA-256 of the same payload the Bitcoin proofs commit to, so both kinds of evidence date one string. Our signature Both commands above check it with --key anchors/ . A head is listed here only when it verifies. That an earlier head is part of a later one GET /api/v2/evidence/consistency returns the RFC 6962 consistency proof between two sizes. Every entry under an earlier head is under every later one, so a later head’s block bounds the earlier entries too. What this does not prove A head in a block existed before that block was mined. That is an upper bound, never the exact moment: the time between an entry being recorded and the next head an authority stamped is our word, and heads are stamped every two hours when the log has moved. Only each day’s newest head goes into Bitcoin, from 18 October 2026; earlier heads have tokens but no block of their own. The time-stamping authorities named here are not qualified eIDAS trust services, so their tokens are evidence of time without the legal presumption Article 41(2) of Regulation (EU) No 910/2014 gives a qualified one ( the specification ). Since 23 October 2026, six staging witnesses of the public witness network cosign the log’s C2SP checkpoints ( /witnesses ); nobody runs our own portable witness yet. None of this says anything about what a record contains: we never see the record, only its salted fingerprint. ## Questions and answers about RiskRouter URL: https://riskrouter.eu/faq Questions and answers Questions and answers about RiskRouter The questions people ask before they write to us, answered in a sentence or two. Each answer links to the page that holds the detail, so nothing here has to be taken on our word. If your question is not here, ask it . What RiskRouter is What is RiskRouter? Evidence infrastructure for anyone who may one day have to show that a record was never changed: advice, credit and KYC files, AI decisions, security logs. A firm seals a salted fingerprint of each record in a log nobody can edit, including us, and anyone can check it later offline; the record itself never leaves the firm. It is built first for software vendors serving regulated finance and for AI platforms, so one integration serves every firm on them. A server-side pricing engine is kept as a worked example for developers. For software vendors · For AI platforms · Industries Is RiskRouter an insurer or an insurance intermediary? Neither. It is not an insurer and, in our own analysis, not an insurance intermediary: it gives no advice, proposes and concludes no contracts, never deals with the policyholder, and is paid a flat software fee. It is not registered with the FSMA. What the FSMA said, and what it did not · Regulatory position Is RiskRouter required by law? No. No law requires RiskRouter or any other product. The law requires the records themselves: IDD, MiFID II, the AI Act, GDPR Article 22, DORA and NIS2 each ask a firm to do something and later to show it did. What RiskRouter adds is a way to show a record was never changed that does not rest on anyone’s word. What the law asks you to record Does RiskRouter sell insurance? No. Nothing on this site is an offer of insurance and no cover is in force. The prices the demo shows come from a simulated rate grid, and no contract can be concluded through it. Regulatory position Is it in production? It is a validation build. The API, the evidence log, the proofs and the verifiers are live and can be used with a sandbox key today; pricing is deterministic mock rules, and there is no customer yet. Status · Usage in numbers Data and privacy What data does RiskRouter receive? For the evidence log, only a salted SHA-256 digest of your record and a short tag such as ai.decision . The record itself, the customer’s name and the salt stay with you. Pricing a quote involves no personal data at all. Data protection · How a digest is made Can an entry be deleted? No. The log is append-only on purpose, which is what makes it evidence. That is why it holds only digests: the record stays in your own systems, where you can erase it when the law requires, and a salted digest on its own says nothing about anyone. Specification Where is data stored? In Frankfurt, inside the EU. An encrypted daily backup of the ledger is kept on GitHub for 90 days, and it never includes contact-form enquiries. Data protection · Trust Is RiskRouter certified (ISO 27001, SOC 2)? No. We hold no certification, and our providers’ certifications do not make us certified. What we offer instead is evidence you can check yourself. Security questionnaire Proof and continuity How do I check a record without trusting RiskRouter? With the open-source verifier, offline: it checks that your record gives the recorded digest, that the entry is in the log, and that the log’s signature holds. Public timestamps and a Bitcoin anchor fix when each head existed. Verify a record · Open-source verifier Can it show that nothing was left out? Yes, for records you put in a numbered chain. The log refuses a position out of turn and signs the list of every position it holds, so a record sealed and later held back shows as a gap. It cannot show that a record was put in a chain at all; that is a policy for you or your supervisor to set. The same chain lets an auditor check a sample drawn from a Bitcoin block hash nobody could choose. Is this all of it? Can I show one field of a record without the rest? Yes, if the record was sealed field by field. You reveal one field’s value and its salt, the verifier checks it against the sealed record, and the other fields stay sealed. Their names are visible; their values are not. Selective disclosure What happens if RiskRouter disappears? The evidence outlives us. Every proof checks offline with the public verifier, the log’s keys and anchors are published, and you can export what you recorded at any time. The engine can also run in your own account. Continuity Who else checks the log? The log publishes signed checkpoints in the format the public witness network cosigns, and anyone can run a witness of it. Since 23 October 2026 each new checkpoint is posted to witnesses of that network (six since 24 October), and the page counts the cosignatures from the published files. Witnesses Are the timestamps qualified under eIDAS? No. Every signed head gets RFC 3161 tokens from two independent time-stamping authorities and a Bitcoin anchor, but those authorities are not qualified, and every page and verifier says so. A qualified authority can be added by contract when a customer needs one; the tooling that requests and checks its tokens is already in place. Specification Is the log a qualified electronic ledger under eIDAS? No. Since Regulation (EU) 2024/1183, the records in a qualified electronic ledger enjoy a presumption of their ordering and integrity (Regulation (EU) No 910/2014, Article 45k). Only a qualified trust service provider can run one, and RiskRouter is run by one person with no legal entity yet. The log’s records carry the weight a court gives any electronic evidence, which Article 45k says may not be denied solely because a ledger is electronic or not qualified. Continuity Getting started How do I start? Take a sandbox key yourself, with no form and no contract, and run the integration kit against it. Integrate How much does it cost? There is no commercial price yet: this is a validation build, and the sandbox is free. The rating matrix the engine prices is public. API reference · Changelog Which languages and systems does it support? Any system that can make an HTTPS request. The integration kit is ready in PHP, C#/.NET, Python and Java using only their standard libraries, and drop-in recorders for Python and Node turn existing decision logs into evidence. No SDK is required. Integrate Can AI agents use it? Yes. Any MCP client can connect to https://api.riskrouter.eu/mcp to read the log and, with a key, record digests; in Claude, it is in the connector directory . Adapters for LangChain, LangGraph and the OpenAI Agents SDK record each agent run as one entry. There is no pricing tool for agents. Connect an AI agent Can we run it ourselves? Yes: the same engine runs in your own cloud account or on your own servers, against your own database, and calls nothing of ours. Running it needs the source and a licence agreed in writing. Continuity Is it open source? Everything you need to check our evidence or build on it is, under the Apache License 2.0: the specification and its test vectors, both verifiers, the conformance suite, the regime packs, the integration and verification-page kits, the portable witness and the drop-in recorders. The service that runs the log is not licensed, and the design means you never have to trust it. Specification · Verifiers · Recorders