What is true today, and what is not.

When the ChainChoice agent names a pick, the answer is saved as a receipt at a public /r/<id> URL. After the write, ChainChoice updates only fields outside its content hash (a recorded outcome, a moderation status, an account binding); an edit to a hashed field would no longer match the hash.

Each receipt carries a sha256 over the canonical block of everything it states, and the hash of the previous receipt in the same anonymous session; a receipt with no session starts no chain.

Methodology version, agent build SHA, model id, prompt version and tool-stack hash are recorded when a receipt is written. They identify what produced the answer. They do not make it reproducible: the answer is model-written text, and nothing re-runs a receipt.

The market figures relevant to a question are stored with a materiality threshold for each. When a receipt is opened, current values are compared with them and moves past the threshold are flagged. The comparison is computed on read and not stored; the receipt itself is not changed.

A build check walks every module the ranking can reach and fails the build if any of them can see affiliate, commission or advertising data. No outside party has audited it.

Versioned methodology with a public changelog. Every weighting and eligibility change bumps the version; receipts pin the version active at write time.

The status line on every page reads "All systems operational" while the oldest of the five live feeds (DEX, layer-2, lending, perpetuals and staking figures) is at most 2 days old, counted in UTC calendar days. Past that it reads "Feeds last refreshed N days ago" in amber. A feed whose refresh fails keeps its last snapshot instead of going blank, so the line reports the oldest one. It is checked when a build ships and again in your browser on the day you read it. It concerns the data pipeline only; how long a dated figure counts as current is a separate rule on the Methodology page.

Snapshots of every cited URL, so an audit does not depend on the public web staying the same. The database has a table for them; nothing writes to it yet, and no date is set.

Approvals, exports and escalations recorded against each receipt. The database has a table for them; nothing writes to it yet, and no date is set.

A new receipt written when a watched figure moves past its threshold, leaving the original as it was. The database has a table for them; nothing writes to it yet, and no date is set.

Not started. We will engage an auditor and build a readiness program once we have a paying institutional pilot to anchor it. Estimated 9–12 months from kickoff to Type II report.

Standard contract surface for EU institutional buyers. Will produce on first pilot request.

Recommendation receipts are hashed, not signed. Signing each with a key whose public half is published would let anyone check a receipt without trusting ChainChoice; it is not built.

A PDF of a receipt with its full trail as an appendix. Not built, and no date is set.

Third-party pen test of the API and receipt surfaces. Will commission ahead of the first signed pilot.