[JacobStephens2]

Summary

GitHub wait unchanged:
  first minute       → ask every 10s
  through minute four → ask every 30s
  thereafter         → ask every 60s
observed change      → restart at 10s
CI poll              → observe PR once; gate re-runs again before writing

Applies to CI, PR mergeability and existing Base fix waits. The existing test override scales the schedule. The manual describes the cadence.

Closes #2138

Evidence

Execution evidence

• Before: cargo test --test ci polling::thirty_minutes -- --nocapture failed: 47 CI polls over 30 simulated minutes.
  After: the same command passed (1 passed; 0 failed), enforcing at most 40 polls and a reported green result.
• Before: cargo test --test ci polling::a_pending -- --nocapture failed: PR observations between pending CI reads, left: 2, right: 1.
  After: the same command passed (1 passed; 0 failed).
• Spec review regression: the reviewer's exact append/run/restore command failed: head move was observed, but the next CI poll waited 1524101us; expected the 250ms shortest step. After retaining the test and including the observed head in backoff, cargo test --test ci polling::an_observed_head_move_resets_the_pending_ci_pause -- --exact --nocapture passed (1 passed; 0 failed). Its final timing uses a 1s shortest step, 6s maximum and 4s assertion to leave several seconds for subprocess work, addressing Standards' deadline finding.
• The final timing version of that exact focused Cargo command also passed (1 passed; 0 failed, 39.35s).
• cargo fmt --check passed. CARGO_BUILD_JOBS=2 scripts/affected-tests main selected site checks and the full Rust suite: site checks passed in 107.10s; the duplicate local Rust run was stopped without reported failures after CI completed successfully on the exact same head, reusing that coverage per docs/testing.md.
• Full CI passed on 9a28b50fbd2d3b8aea7ead73d2a74c1b84c9d3c8, including all four Rust partitions, their coverage gate, formatting/typechecking and site/Console checks.

Visual evidence

Manual waiting cadence, desktop 1440×900:

Manual waiting cadence, mobile 390×844:

Test seams

• The Command binary against fake GitHub, with the existing shortened poll interval for long waits.
• The poll module's schedule at explicit elapsed durations.

Review coverage

Reviewed against refs/remotes/origin/main, pinned at 943166dea1ef448c2a50e345bedc6b3ef6257bd4. Standards raised one finding and Spec raised one; both were addressed.

Both axes read all eleven changed source/test/documentation files. site/docs/index.html was intentionally excluded from source reading by both axes: docs/guide.md and site/generate-docs.mjs were reviewed, npm run docs:check --prefix site passed, and the rendered desktop/mobile captures were inspected. No other changed files were left unread.

Unaddressed findings

Standards

None.

Spec

None.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Blast Radius: Commands

Unchanged GitHub waits can take up to sixty seconds to notice completion. Stop and interruption keep the existing 100ms sleep checks.

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