[JacobStephens2]

Summary

release lock + origin reservation
  discover unfinished release or pinned content
    human blocker -> quiet, deduplicated Pass skip
    exact bump -> resume; failed unpublished tag -> recover
  optional Experiment -> terminal override or unattended refusal
  summary -> optional terminal accept / editor / cancel
  bump PR checks -> at most one targeted Fix -> exact-tree validation
  publish tag -> verify GitHub Release -> merge bump PR

Adds --with-experiment to the CLI, admin-scoped start API and Console. --review is a reasoned CLI-only parity exclusion. Published Releases stay final.

Closes #2159

Evidence

Execution evidence

• Before: cargo test --test release native::native_pass_quietly_skips_human_blocked_superseded_cuts -- --exact --nocapture exited 101: superseded release 0.2.0 requires human review; 0 passed; 1 failed.
  After: the same command passes: native_pass_quietly_skips_human_blocked_superseded_cuts ... ok; 1 passed; 0 failed. The retained regression covers draft and foreign-commit blockers, quiet repeated Passes, preserved refs and deduplicated Activity log entries.
• Before: node --test --test-name-pattern='Release start requires admin' console/tests/start.test.cjs failed with locator.check: Timeout 5000ms exceeded waiting for Run release Experiment.
  After: the same command passed; the full Start browser file also passed all 16 tests, including desktop and mobile requests.
• The real Experiment-engine release regression passes: it refuses an unattended inconclusive Verdict, resumes the pinned commit after main moves without repeating Trials, and permits a terminal override.
• CLI parity: 3 passed. Admin start API: 3 passed. cargo clippy --all-targets --locked -- -D warnings: passed.
• The local selector passed site checks (69 tests), Console checks (207 tests), and cargo build --locked. Full Rust validation passed in all four CI partitions on exact head 53cbcda3 (CI run). Following the instruction to reuse successful CI, the duplicate local Rust run was stopped after CI completed; its selector exit reflects that deliberate interruption. No local assertion failure was reported before cancellation.

Test seams: the release CLI against real Git and fake GitHub/Harness executables; the admin start HTTP interface; the Console Start form and confirmation flow. No tests target private release helpers.

Visual evidence

Console Start Release run — 1440×900

Console Start Release run — 390×844

Expanded release manual — 1440×900

Expanded release manual — 390×844

Expanded How it works release flow — 1440×900

Expanded How it works release flow — 390×844

Expanded architecture release flow — 1440×900

Expanded architecture release flow — 390×844

Review coverage

Standards: 3 findings, all addressed. Spec: 1 finding, reproduced with the reviewer's exact regression and fixed. Both independent reviewers read all 31 changed files, including generated documentation hunks. Changed files left unread: none. Generated-output exclusions: none.

Unaddressed findings

Standards

None. Typed release PR observations, named unfinished-release fields and shared PR identity interpretation address all three design findings.

Spec

None. Superseded release blockers now return a quiet Pass skip; the reviewer's regression is retained.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: one-way

Running the release command can publish permanent artifacts. Recovery only moves a tag after verifying that its GitHub Release is unpublished, and uses expected-ref leases.

Blast Radius: releases

Built with codex codex-cli 0.167.1 · gpt-6.1-sol · xhigh