Leaks
Accuracy ledger
Every claim ENTERED IN THIS LEDGER and what became of it. Append-only: a claim line is written once and a resolution line is appended later naming it, so the outcome cannot be edited into the record after the fact. Enforced by .github/workflows/append-only.yml.
It is not a scorecard for the whole desk, and saying so is the difference between a ledger and a boast. A claim here is a PREDICTION that can turn out to be right or wrong, and it scores a FIELD rather than a company: whether OpenRouter's catalog expiration_date predicts the catalog's own behaviour. Nothing here says a lab missed a deadline. A retirement floor cannot be scored this way, because those dates are in a provider's own API namespace and joining anthropic/claude-opus-4.1 to claude-opus-4-1-20250805 is the guess this archive refuses. A derived item on the desk describes a stored artifact, predicts nothing, and is never scored here; the desk currently holds 17 items.
- Recorded claims
- 3
- Confirmed
- 0
- Refuted
- 1
- Open
- 2
- Accuracy over resolved claims
- 0%
| Recorded | Claim | Tier | Outcome | Resolved | Artifact |
|---|---|---|---|---|---|
| 2026-08-30 | OpenRouter's catalog listed an expiration_date of "2026-08-31" for "moonshotai/kimi-k2.5". Scored as: "moonshotai/kimi-k2.5" is absent from the catalog after that date. Still listed in the catalog captured at 2026-09-01T10:12:17.000Z (origin_date), after an expiration_date of "2026-08-31". |
confirmed-artifact | refuted | 2026-09-01 | artifact |
| 2026-09-01 | OpenRouter's catalog listed an expiration_date of "2026-09-04" for "nex-agi/nex-n2-pro". Scored as: "nex-agi/nex-n2-pro" is absent from the catalog after that date. | confirmed-artifact | open | not yet | artifact |
| 2026-09-01 | OpenRouter's catalog listed an expiration_date of "2026-09-04" for "nex-agi/nex-n2-mini". Scored as: "nex-agi/nex-n2-mini" is absent from the catalog after that date. | confirmed-artifact | open | not yet | artifact |