perf(ci): avoid duplicate runners for offline auto-merge checks - #607

🤖 Generated by the Agentic Engineer

Pre-submission correctness and simplification review is clean at ec064d99749aa6bcacda4342918bdfb62a6f948d, including an independent read-only assessment of the final change. The only job admission change is !inputs.dry-run: true offline calls retain their dedicated fixture runner; omitted/false inputs retain the required eligibility reporter. Trust classification, privileged-job predicates, permissions, triggers and mutation serialization are unchanged.

The guard failed against the original redundant admission condition before the workflow changed. All 35 independent dry-run mutation controls, author/trust/disarm tests, the actual safe no-op reporter, native full-fixture controls and all 14 maintenance checks pass. Catalogue credential analysis, CI selection/reducer checks, README parity and Bash lint pass. GitHub verifies the signed commit as valid.

Hosted user-path verification remains required before promotion: all three offline callers must execute their fixtures successfully while allocating no eligibility runners; direct/input-free eligibility must still execute successfully using the candidate workflow. Its new public source-SHA receipt must bind the actual workflow revision to the reviewed workflow blob. This is not yet a delivered runner reduction, a whole-CI latency result, an API-pressure result or a production fleet-capacity result.

🤖 Generated by the Agentic Engineer

Review routing at ec064d99749aa6bcacda4342918bdfb62a6f948d: a fresh read of the authenticated CodeRabbit quota artifact confirms the included public-repository allowance is exhausted until 2026-10-15 18:53:10 UTC, calculated from its 18:23:10 UTC provider event and stated 30-minute reset. That repository-scoped limit still applies at this request. The mandated review-health check agrees, and Codex is serving. Advancing to Codex without spending a redundant CodeRabbit request or paid capacity. This is availability evidence, not a review verdict.

🤖 Generated by the Agentic Engineer

@codex review

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Codex Review: Didn't find any major issues. What shall we delve into next?

Reviewed commit: ec064d9974

Code coverage overview

Languages: Go, C#

Go / code-coverage/test

C# / code-coverage/dotnet

Go / code-coverage/go

ℹ️ Calculations represent line coverage. Results may change as calculations are updated.

_{Updated October 15, 2026 20:39 UTC}

🤖 Generated by the Agentic Engineer

Hosted validation at ec064d99749aa6bcacda4342918bdfb62a6f948d now proves the bounded allocation change. The complete native job inventories reconcile their advertised counts: baseline run 38775179174 allocated six runners across the three offline callers; candidate run 38788767226 allocated three. All three retained fixture jobs and their five decision suites passed. All three duplicate eligibility jobs skipped without a runner or steps, and all six write-capable jobs skipped in both runs. This does not establish whole-CI latency, API savings or fleet capacity.

The separate input-free run 38788766783 completed its eligibility reporter successfully. Its executed log records source e4467778d5331c6f7697cbc81b3c10ffc6b12ff3; the provider resolves that workflow to blob 19e93141cc46bd2582754d7cd281c51185fb2c10, exactly matching this PR's workflow. Its fixture and write-capable jobs skipped.

That candidate also exposed a stale mutation control: forcing a downstream condition to true leaves its skipped dependency in effect. The signed follow-up f9ff40fdac6c2c395e7d33138d0ae273f29336b0 changes that control to always(), which restores a genuinely reachable writer. The full 45-control credential suite, actor gate fixtures and ShellCheck pass locally; production workflow bytes are unchanged. Fresh hosted CI and a current-head review are required before merge.

🤖 Generated by the Agentic Engineer

@coderabbitai review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

🤖 Generated by the Agentic Engineer

@codex review

Codex Review: Didn't find any major issues. You're on a roll.

Reviewed commit: f9ff40fdac

🤖 Generated by the Agentic Engineer

Current-head hosted evaluation at f9ff40fdac6c2c395e7d33138d0ae273f29336b0 passes the runner-allocation contract. Complete job graphs for baseline 38775179174 and candidate 38796618847, each attempt 1, show six versus three runner allocations across the three offline callers. All three fixture jobs and their five decision steps pass; the three duplicate eligibility jobs allocate no runner. Privileged jobs remain skipped in both graphs.

The ordinary input-free path also passes in 38796618137, attempt 1, at this head: eligibility succeeds on one runner; the fixture and both privileged jobs are skipped without runners. Its executed source receipt identifies 311c4a873eda763aba5d7d201c41287c9641f2a0; the workflow blob matches the reviewed head exactly.

This demonstrates runner allocation and preservation of the input-free reporter, without claiming whole-CI latency, API-quota or fleet-capacity improvements. Full required CI remains pending.

🤖 Generated by the Agentic Engineer

Ready at f9ff40fdac6c2c395e7d33138d0ae273f29336b0. Complete current-head CI 38796618847 and all readable checks pass. The native acceptance record above demonstrates six versus three allocations across the three offline callers, preserved fixtures and skipped privileged paths, plus the ordinary input-free reporter at this exact head. No whole-CI latency or fleet-capacity result is asserted.

The authenticated current-head Codex review is green. Complete comment/review/thread reads show no outstanding findings or maintainer direction, and GitHub reports CLEAN. Every readable required gate passes; the configured code-quality rule has no independently readable verdict, so ordinary protected merge remains authoritative. Promoting now; the promotion-triggered check must settle before merge.