Benchmark-quality register

We measure the instruments too.

Everyone argues about which model tops which leaderboard. Almost nobody asks whether the leaderboard is built well enough to settle the argument. This register asks — with deterministic predicates, answered from a benchmark's own public artifacts, each one carrying the URL it was read from and the date it was read.

No language model judged anything here. A register that graded other benchmarks on scoring transparency by asking a model for its opinion would fail its own predicate on the first row.

Record typemeasured current stateSolicitednoSubject participationnoneAccesspublic artifacts onlyEndorsementnoneCertificationnot a certificationModel judgmentnone
The signed JSON What those benchmarks report →Our own losses →

We are not on this register, and we cannot be

Council of AI does not assess its own instruments on this register. The exclusion is enforced in the code that builds the payload, not merely promised in prose.

The register endpoint could not be read (SyntaxError: Unexpected token '<', "<!doctype "... is not valid JSON). Nothing is shown rather than a cached copy — a stale register that looked live would be the exact failure this page exists to refuse.