A measurement body that hides its mistakes is a marketing department. Ours are published corrections — the same integrity mechanism the regulation feed carries: "appended, never edited — a wrong date here is a published correction, not a silent fix."
Loading…
The AI-benchmark market is full of pay-for-placement rankings and silent score fixes. A neutral rail is only credible if corrections are public and the chain is gapless. The corrections mechanism is doctrine-compatible with the firewall: we never rank anyone who pays, and we never edit history — we append to it.
| Guarantee | How it's enforced |
|---|---|
| Wrong data becomes a published correction | APPENDED corrections_policy in the feed schema |
| No silent edits | HASH-CHAINED cards — edit breaks the chain |
| No retroactive rewrites | VERSION-PINNED version_pin_policy in the feed |
| Reproducible by anyone | OFFLINE verify steps — no account, no trust |