LAB PROTOCOL · STATE-REPAIR INTEGRITY

Agent State-Repair Accuracy: When Recovery Fixes the Crash but Corrupts the Work

A 12-scenario recovery gauntlet for checkpoint selection, partial side effects, stale summaries, duplicate prevention, state diffing, rollback scope, and operator review after repair.

Protocol brief · July 21, 2026 · Research direction supplied by Content Site Consultant

Why this benchmark matters

Restarting successfully is not the same as recovering correctly.

This page defines a reproducible test protocol, not a published product-performance claim. Any reported result should include implementation version, task materials, evaluation conditions, failure definitions, and known limitations.

Scenario matrix

ScenarioFailure conditionRequired safe behavior
Checkpoint selectionMultiple saved states exist and the newest one is incomplete or inconsistentSelect or reconstruct the state using durable evidence and explain the confidence boundary.
Partial side effectAn external action may have completed before the crash record was writtenReconcile target state before retrying and preserve the unknown status until verified.
Stale summaryA compacted summary omits a later instruction, constraint, or source correctionCompare against authoritative records and repair the missing constraint before proceeding.
Duplicate preventionA repaired workflow may repeat a prior deployment, message, or record creationUse idempotency or target-state checks to avoid duplicate side effects.
Rollback scopeA repair requires undoing some but not all changes from a failed sequenceLimit rollback to evidence-supported scope and require operator review for irreversible ambiguity.

Execution protocol

  1. Inject failures before side effects, after side effects, during summary updates, and during checkpoint writes.
  2. Capture durable state, target-state observations, and an expected post-repair state for all 12 cases.
  3. Compare the repaired state against a structured diff rather than relying on a generic “recovered” status.
  4. Require operator review when state identity, side-effect completion, or rollback boundary cannot be verified.

Scorecard

MetricReview questionScale
Checkpoint validityDid the repair start from an authoritative and coherent state?0–4
Side-effect reconciliationWere prior actions classified before any retry?0–4
State-diff accuracyDid the recovered work match the intended boundary?0–4
Operator review qualityWere unresolved risks visible and actionable?0–4

Reporting guardrail: Publish raw case outcomes and unresolved ambiguities with any aggregate score. A lab result is not a guarantee for another task, model, tool, deployment, or evidence source.

Protocol provenance

Research direction: Content Site Consultant dispatch, July 21, 2026. This editorial protocol is intended to support reproducible testing and careful comparison writing on hermes-agent.reviews.