v0.1.0-beta.2 tagged — shadow-evaluation beta, pilot invite open

[daichunghy]

PatchGate checks whether a pull request has the trusted base policy, commit-bound check evidence, ownership and a qualified human gate before it is represented as ready for maintainer review. It is deterministic, evidence-bound, and deliberately narrow: it does not detect AI authorship or replace human judgment.

v0.1.0-beta.2 is the current pre-release. It supersedes beta.1, which a live maintainer smoke found unreadable on real runners (the GitHub Actions runner exports dashed input names like INPUT_GITHUB-TOKEN; the parser now reads runner-native names first). The findings record is in the repository: docs/reviews/2026-08-22-live-smoke-findings.md.

Honest boundaries, so nobody is surprised:

• shadow evaluation only — fail-on: never, no required check, no merge blocking;
• merge_group inputs return an explicit evidence_missing, not a guess;
• complete native-control snapshots need a PAT or GitHub App token with administration: read (the workflow GITHUB_TOKEN cannot be granted that permission);
• no external pilot, adoption or production claim exists yet.

If you maintain a public repository with protected branches and CODEOWNERS and want to try a consented, non-blocking shadow observation, the pilot request, shadow runbook and Action usage guide describe the exact scope, permissions and rollback. A technically informed critique of the trust-boundary claims is equally welcome — it needs no installation.