[eliasruntime]

External feedback identified a concrete importer-count question, but trying Proofline still required a maintainer to invent a task and configure all evidence first. Add a standalone offline probe for the reported Driftproof behavior and an optional sidecar that prepares two explicit checks without replacing existing project configuration.

At pinned upstream revision 871c8c6a82b57934e42a8e4f4b1bcec5f49e08f8, authored inputs with no judge-sample count produce counts 1 and 2. Declaration preservation passes; the newly discussed count contract fails. The already-published Proofline 0.2.0-rc.1 CLI records one verified control and one failed criterion. Upstream tracked files were not changed. The standalone probe is usable without Proofline, and both the README and redacted record separate this maintainer execution from independent use, the upstream full gate, model runs and net benefit.

Validation: 59 local tests, lint and the 42-artifact release gate passed. Added regression coverage checks the positive/negative count controls, reward retention, existing-file refusal, spaced paths, source invalidation and a subsequent corrected observation. Hosted CI will cover the same source and installed CLI matrix. There is no runtime API change, release retagging or physical-Mac claim.

[eliasruntime]

All nine Windows/macOS/Ubuntu × Node 20/22/24 jobs passed on 62d2c0f, including package installation: https://github.com/codex-improvement-lab/codex-proofline/actions/runs/35533245785 . The suite has 59 tests.

Those CI jobs validate Proofline and the new probe/sidecar regression tests using synthetic importer controls. The actual pinned Driftproof reproduction is separately recorded as a maintainer-run Windows observation with the published Proofline 0.2.0-rc.1 CLI. No external-user, whole-task-benefit, or upstream-full-gate result is implied.