Stable-release and adoption gates

GitHub prereleases are intentional while the reducer has strong controlled evidence but little independent field evidence. A non-prerelease should not be used as a substitute for real adoption.

Required before the first stable GitHub release

• CI passes on Windows, macOS, and Linux with every supported Node.js line.
• The weekly Harness next profile canary is green.
• At least three independent, privacy-reviewed field reports exercise real profiles. Useful outcomes include isolating one or more community bundles or correctly returning a baseline/scope failure.
• No report shows source-profile mutation or an unredacted secret/path leak.
• The public install asset, checksum, clean install, and uninstall are verified.
• The official Harness maintainer discussion has been posted and linked, even if maintainers have not yet selected an upstream direction.

Evidence, not vanity metrics

Stars, clone counts, release downloads, and catalog inclusion are discovery signals. They are not field reports: CI, release smoke tests, and maintainer dogfooding can generate the same events. Only a report from an independent real profile counts toward the three-report gate.

Feature gates

• Do not add broader order/prefix reduction until a real report demonstrates that set reduction loses the failure.
• Do not add automatic repair or quarantine; a minimal set is evidence, not a safe-removal decision.
• Prefer an upstream profile-inspection/reporting seam over additional private manifest assumptions.

Field reports use the repository's Real Harness field report issue form and must be manually reviewed for privacy before submission.