Audit an analysis

Determine how far a completed analysis can be trusted and what its session traces teach us about the analysis process. Produce evidence that helps maintainers distinguish local mistakes from recurring weaknesses and successful safeguards—not a change proposal justified by one run.

Choose the inspection approach. Make consequential findings traceable to evidence, distinguish observations from recommendations, and state what remains unknown. Compare analysis timings across available runs to identify what drives cost or delay, keeping differences in scope, worker profile, and parallelism visible. Distinguish wall-clock duration from summed worker time; one run alone does not establish a performance trend. A partial audit is useful if its limits are clear.

Save the report under reports/ and add it to the workshop's report list. Include run-id, method-commit, source-revision, and your own reporter-profile (harness, model, reasoning effort; unknown where unavailable), distinct from the audited worker profile. Leave the method, retained analysis, and run state unchanged.