Project / Experimental open source
Making release evidence reviewable.
Release Evidence examines one local software repository and helps a technical owner assemble the facts and human judgments behind a release or handover decision.
- Project
- Release Evidence
- Status
- Experimental source preview
- Published
- October 2026
- Stack
- Go, TypeScript, local web UI
The problem
The decision is one moment. Its evidence is everywhere.
Build definitions, tests, ownership, deployment notes, recovery guidance, and security reporting often live in different files and different people's heads. The accountable person still has to work out what was observed, what was not found, what remains unknown, and who owns the gap.
The working loop
Scan. Review. Preserve.
- 01 / Scan
Observe a bounded repository.
Deterministic checks inspect supported local files, cite what they find, and explain the search behind evidence they do not observe. The default path makes no network request.
- 02 / Review
Keep judgment visibly human.
A loopback-only interface separates observed, not observed, unknown, and not applicable states. The reviewer can add context, ownership, rationale, and follow-up actions.
- 03 / Preserve
Leave a traceable packet.
Review state is saved separately from the scanned repository, with machine observations and manual assessment kept distinct for the person making the decision.
Boundaries by design
Useful evidence without a fictional score.
Local by default
Repository contents stay on the operator's machine; telemetry and hosted accounts are not required.
Observation, not assurance
Finding a file does not prove a practice is effective, and missing evidence is not proof that it does not exist.
No automated approval
The tool does not produce a readiness score, certify compliance, or decide that a release is safe.
Hostile input assumed
The scanner does not execute repository content, follows bounded readers, and refuses escaping links.
What exists today
Public code. Working workflow. Open question.
The scanner and local review slice are public under Apache-2.0 and run from source. An internal founder baseline exposed workflow problems; focused corrections now pass the automated suite.
What this does not prove: independent demand, external usability, a safe release decision, or production support. There is no stable binary release or commercial offering yet.