Closed-loop effectiveness

lessonweaver can aggregate runtime usage events into conservative effectiveness reports. These reports do not prove causality; they summarize observed signals so maintainers know which reviewed skills deserve another governance pass.

Use SkillEffectivenessReporter with a registry:

from lessonweaver import FileSystemRegistry, SkillEffectivenessReporter

registry = FileSystemRegistry()
reports = SkillEffectivenessReporter().report(registry)
for report in reports:
    print(report.to_dict())

Reports distinguish:

• improvement — positive graded usage evidence outweighs failures;
• repeated_failure — negative graded outcomes suggest the lesson still misses the failure pattern;
• possible_regression — a negative usage outcome or note explicitly mentions a regression after the skill loaded;
• staleness — an active skill has no usage evidence;
• insufficient_evidence — the available events are not enough to recommend a stronger action.

Recommendations are intentionally review-oriented: keep, revise, deprecate_or_revise, or review. Operators can feed these reports into the existing stale-report and cleanup workflows instead of trusting activated lessons forever.