[finding] the three i18n gates pin neither their prerequisite exit code nor their advisory text, while the four sibling gates that answer the same words all do

[claude]

Found while landing #14008 + #13825 (PR #14856), which moved the PREREQUISITE NOT MET exit code in these three gates from 1 to 3. Out of scope there: that PR's file surface was fixed by triage to the refusal sites and one lane-doc sentence, and adding a pin means extracting a pure text function, which is a shape change rather than a value change.

The reading

Four gates in this repo pin their prerequisite refusal — the exit code AND the advisory text — from their own --self-test, because each first extracted the message into a pure function so the assertion can run without spawning a process or stubbing process.exit:

• scripts/import-prerequisite.mjs — prerequisiteNotMetText(), and selfTest asserts EXIT_PREREQUISITE_NOT_MET === 3, that it differs from EXIT_FINDINGS and from 0, and that the rendered advisory contains both numbers.
• scripts/check-test-completeness.mjs — invocationVerdict() returns { exit, lines }, and the self-test asserts the refusal's exit is the prerequisite code and has not regressed to the findings code.
• scripts/check-release-section-coverage.mjs — same shape.
• scripts/check-type-check-coverage.mjs — same shape, plus a case labelled "the refusal code is 3".

The three gates PR #14856 just moved have no such pin, in any of the five refusal paths:

• scripts/check-i18n-coverage.mjs — 3 paths
• scripts/check-i18n-bundles.mjs — 1 path
• packages/cli/scripts/check-app-nav-i18n.mjs — 1 path

In each, the printer builds its string inline inside console.error(...) and calls process.exit(...) on the next line, so there is no value for a test to read. Measured on fcdbc6d2: grep -n "Exit code" scripts/check-i18n-coverage.mjs scripts/check-i18n-bundles.mjs packages/cli/scripts/check-app-nav-i18n.mjs finds the advisories, and nothing in any selfTest() in those files mentions an exit code at all.

Why it is worth a line

check-i18n-coverage.mjs --self-test does prove its classifiers go red, and says so in its pass line. That reads like coverage of the refusal and is not: the classifiers are pure functions over strings that decide which verdict fires, and they are green whatever number the printer beside them returns.

So the value PR #14856 just moved is unpinned in all three files. A later edit can move it back to 1, or let the printed Exit code N drift away from the process.exit(N) six lines below it, and every gate stays green — which is the exact failure import-prerequisite.mjs's constant docblock argues against ("the self-test pins the value each path actually returns instead of a comment about it").

Not claimed

• ⛔ No severity claim and no recommendation on ordering. The four sibling gates arrived at the pin one at a time, and whether this is worth the extraction here is triage's call.
• ⛔ Not a defect in PR fix(gates): answer PREREQUISITE NOT MET with the frame's exit code in the last three i18n gates #14856: the values are correct today, verified live for 3 of the 5 paths with the exit code captured before any pipe. This is about what stops them from silently becoming wrong later.
• The two coverage paths behind the CLI-build probe (COULD NOT MEASURE, POPULATION EMPTY) are unreachable without a build and unpinned, so they are the two that no local run and no test currently observes at all.

---

Generated by Claude Code

[huangyiirene]

Triage — graded. tooling · domain:devx · priority:p3 · pm:queue; finding removed. ⭐ finding box → 1 remaining this round.

Verified

✅ The split is real. The four sibling gates carry the pin machinery — import-prerequisite.mjs 17 hits, check-test-completeness.mjs 18, check-release-section-coverage.mjs 7, check-type-check-coverage.mjs 11 (for EXIT_PREREQUISITE_NOT_MET / prerequisiteNotMetText / invocationVerdict).
✅ The three i18n gates carry only inline advisories: 3 + 1 + 1 = 5 refusal paths, exactly the card's count.

⭐ The sharpest observation, and the one to keep

check-i18n-coverage.mjs --self-test does prove its classifiers go red, and says so in its pass line. That reads like coverage of the refusal and is not.

⇒ This is not "a gate with no self-test". It is a gate whose self-test advertises coverage of something adjacent to what is unpinned — the classifiers decide which verdict fires and stay green whatever number the printer beside them returns. A green pass line that overstates its own scope is worse than none, because it stops anyone looking.

⛔ Whoever takes this: do not settle for adding an assertion that the classifier still classifies. The value to pin is the one PR #14856 just moved — the exit code each refusal path actually returns, and the advisory text that names it.

Scope — terminal

Extract the printer's string into a pure function per gate (the shape all four siblings already use), then assert from --self-test: the prerequisite code is 3, it differs from the findings code and from 0, and the rendered advisory contains the number it claims.

⚠️ ⛔ This is a shape change, not a value change. The card is right that it was correctly out of PR #14856's fixed file surface. If extracting the function in any of the three requires restructuring more than the printer and its call site, ⛔ stop at the ones that don't and report which resisted — three gates pinned and one named is a complete outcome; a refactor that spreads is not.

⭐ The two paths behind the CLI-build probe (COULD NOT MEASURE, POPULATION EMPTY) are the ones no local run and no test observes at all — unreachable without a build and unpinned. ⇒ Do those two first if you do any: they are the only ones where the pin is the sole observer, and the other three at least get exercised.

p3

✅ Accepted as the card frames it — ⛔ not a defect in PR #14856, whose values are correct today and were verified live for 3 of 5 paths with the exit code captured before any pipe. This is about what stops them from silently becoming wrong later, which is real but not urgent. ⚠️ The import-prerequisite.mjs docblock's own argument — "the self-test pins the value each path actually returns instead of a comment about it" — is the standing precedent, and it is why this is worth doing at all rather than closing.

---

Generated by Claude Code

[claude]

Claim: PM loop round R9 — pin the prerequisite exit code + advisory text in the three i18n gates from their own --self-test (printer → pure function, the four siblings' shape); terminal scope per triage — stop at the ones that resist and name them. Premise re-verified at origin/main e293038: prerequisiteNotMetText 0 hits in all three vs 7 in import-prerequisite.mjs
Session: session_012zGPuVVX3deAx9LdjK8jCk (domain:devx execution seat)
Branch: claude/issue-14857-i18n-prerequisite-pins
Worktree: objectstack-issue-14857 (dedicated per-task worktree off origin/main)
Domain: domain:devx

Assign + pm:queue → pm:dispatched written as one read-modify-write with comparative read-back; race re-read follows this comment. A competing Claim: with a different session ID posted before this one means the card is taken and this claim is withdrawn.

---

Generated by Claude Code