|
| 1 | +# 003 — Audit record |
| 2 | + |
| 3 | +Seven review rounds over this unit, two independent reviewers. Recorded because |
| 4 | +the corrections are more instructive than the plan, and because a unit that |
| 5 | +claims "every finding was verified" should show what verification cost. |
| 6 | + |
| 7 | +## Rounds |
| 8 | + |
| 9 | +| Round | Reviewer | Verdict | Findings | |
| 10 | +|---|---|---|---| |
| 11 | +| r1 | A | FAIL | 6 blockers, 4 citation defects | |
| 12 | +| r2 | A | FAIL | 5 blockers | |
| 13 | +| r3 | A | NEAR-PASS | 2 | |
| 14 | +| r4 | A | (inconclusive) | verdict lost — reviewer closed before the hook recorded it | |
| 15 | +| r5 | B (fresh) | FAIL | 3 blockers, 3 citation corrections | |
| 16 | +| r6 | B | NEAR-PASS | 1 citation defect | |
| 17 | +| r7 | B | PASS | none | |
| 18 | + |
| 19 | +Reviewer B was dispatched with no prior context and explicitly told not to |
| 20 | +assume reviewer A had been thorough. It found three blockers A had passed over, |
| 21 | +including one that would have shipped a false claim about a security control. |
| 22 | + |
| 23 | +## Corrections worth remembering |
| 24 | + |
| 25 | +**A verifier that could not verify.** `031` claimed `privacy:scan` enforced the |
| 26 | +fixed-literal publisher label. It does not — `scripts/privacy-scan.ts:187` is a |
| 27 | +textual scanner over file content and cannot see that a runtime value was |
| 28 | +path-derived. The fix was a closed union type so the constraint fails |
| 29 | +`typecheck` instead. This is the most valuable catch in the seven rounds: the |
| 30 | +plan named a guard that would have passed while the invariant it claimed to |
| 31 | +protect was violated. |
| 32 | + |
| 33 | +**A CI assertion nobody could implement.** `031` also said CI would assert the |
| 34 | +counters stayed zero across the Windows suite. The counters are process-local |
| 35 | +and the suite runs across four sharded runners in many short-lived processes. |
| 36 | +The claim was withdrawn rather than reworded — an instruction that cannot be |
| 37 | +followed is worse than an admitted gap. |
| 38 | + |
| 39 | +**A test verifying the wrong thing.** `051` claimed it could verify `050`'s |
| 40 | +backoff by reverting `050`. Reverting would leave a fixed five-second loop that |
| 41 | +still relaunches, still yields a new PID, still restores health — the test would |
| 42 | +pass either way. Now stated plainly, with backoff verified separately by |
| 43 | +asserting on generated script text. |
| 44 | + |
| 45 | +**Batch arithmetic that fails at runtime.** `050` advised converting `%TIME%` |
| 46 | +with `set /a`. `set /a` reads a leading zero as octal, so `08` and `09` are hard |
| 47 | +errors — confirmed directly: |
| 48 | + |
| 49 | +```text |
| 50 | +C:\> set /a a=08 |
| 51 | +Invalid number. Numeric constants are either decimal (17), |
| 52 | +hexadecimal (0x11), or octal (021). |
| 53 | +``` |
| 54 | + |
| 55 | +Four traps documented in the end: octal, space padding, midnight wrap, delayed |
| 56 | +expansion. |
| 57 | + |
| 58 | +**A job that did not test what it claimed.** `080`'s "self-update end to end" |
| 59 | +used a locally packed tarball, but `ocx update` resolves its target from the |
| 60 | +registry (`src/update/index.ts:167`) and installs a resolved version (`:106`). |
| 61 | +There is no injection seam, so the real command was never exercised. Renamed to |
| 62 | +a package replacement smoke, which is still worth having. |
| 63 | + |
| 64 | +**A gate that does not exist.** `060` promised Windows would block merges. `dev` |
| 65 | +has no branch protection (`MAINTAINERS.md:121`, `:125`). Stage 3 is now a |
| 66 | +convention gate; stage 4 is the real one because `release.yml` reads run |
| 67 | +conclusions directly. |
| 68 | + |
| 69 | +**Sequencing invented after the fact.** `002` originally claimed a long |
| 70 | +dependency chain. Only two links were structural. One was backwards. |
| 71 | + |
| 72 | +**Six citation defects.** `job.ts:1381`→`:1383`, `ci.yml:771`→`:769-772`, |
| 73 | +`release.yml:224-234`→`:224-241`, `service.ts:2330`→`:2340-2341`/`:2350-2355`, |
| 74 | +`config.ts:3937`→`:3942`, and a missing `chmodSync` site the `040` seed list had |
| 75 | +skipped entirely — which is why `040` now says to re-derive the list rather than |
| 76 | +trust it. |
| 77 | + |
| 78 | +## Two claims withdrawn |
| 79 | + |
| 80 | +"Every release to date ran zero Windows tests" was false. Releases touching |
| 81 | +`src/service.ts` and a few other paths separately require a green |
| 82 | +`service-lifecycle.yml` (`release.yml:224-241`), which includes a Windows job. |
| 83 | +The defensible claim is narrower: the release preflight does not require the |
| 84 | +Windows *suite*. |
| 85 | + |
| 86 | +`src/service.ts:1983` was cited as evidence that ACLs are authoritative for |
| 87 | +credential writers. It says so about an elevation staging directory. Evidence |
| 88 | +for the principle, not for any writer's coverage. |
| 89 | + |
| 90 | +## What this says about the unit |
| 91 | + |
| 92 | +Sixteen findings against a document that had already been written carefully. |
| 93 | +Every one was reproduced against the tree before being acted on, and two of the |
| 94 | +reviewer's own line numbers were off in the other direction and corrected back. |
| 95 | + |
| 96 | +The rate at which confident-sounding planning prose turns out to be wrong is the |
| 97 | +argument for `060`. A plan gets seven adversarial rounds; a merge to `dev` |
| 98 | +currently gets no Windows execution at all. |
0 commit comments