---
name: review-evidence
description: Compare a current claim with dated observations and relevant heritage without turning summaries into proof.
---
# Review evidence and heritage

Use when evaluating readiness, learning from a recurring failure, resolving a contradiction, or choosing between proposals.

1. Write the claim as a concrete observable outcome.
2. Read the relevant current capsule and one matching entry from /heritage/timeline.json or /heritage/lessons.json.
3. Keep facts typed: a plan, configuration, process, submission, result, independent verification and useful outcome are different observations.
4. Compare source time, observation time, evidence class, scope and source selector. A private receipt summarized publicly is a reported source, not a receipt you personally inspected.
5. Preserve conflicting accounts. Check whether they concern different times, components or scopes before treating one as a correction.
6. Name one observation that would disprove the preferred explanation. A public reader can propose a test but cannot claim to have executed private verification.
7. Check whether an existing successful mechanism can address the first failing step. More components or more model calls do not by themselves improve the outcome.

Return: claim; support; strongest counterevidence; confidence expressed in plain language; what remains unknown; one falsifier or next verification step.

Example: a feasible numerical candidate that fails to beat the incumbent is a valid negative result. Preserve it as evidence and keep the incumbent; do not describe it as improvement.

Sources: public adaptations of H3 failure lessons and H4 operational heritage. The controls described in heritage may no longer be enforced; check current evidence.