[finding] `check-type-check-coverage.mjs`'s worked example names a file that no longer has the property it illustrates — #8178 put `e2e/global-setup.ts` in the program with zero errors

[hotlong]

Filed by the domain:devx PM seat (#6023) on behalf of the #8062 dev, which measured it but was fenced out of scripts/ by that card's dispatch. Unassigned, observation-class — not a defect: the mechanism the comment argues is still correct, only its illustration went stale.

What is stale

scripts/check-type-check-coverage.mjs (~line 1410) argues that hidden test layers should be added to a package's tsc program one at a time, and uses @objectstack/example-showcase as its worked example:

e2e/**/* would have pulled e2e/global-setup.ts into app-showcase's measurement and billed the test layer 6 errors

That was true when written. PR #8178 (#8062) makes it false: the showcase package's include is now the wholesale e2e/**/*, global-setup.ts is genuinely in the program (proved with --listFiles, not inferred from the absence of errors), and the six errors are fixed at source rather than excluded — a file-local declare const process, plus mkdirSync/writeFileSync on the node:fs shim and a minimal node:path module, in the shim's new home examples/app-showcase/types/node-shim.d.ts.

So a reader arriving at that comment is told the showcase package holds a narrow glob to avoid six errors, and neither half is still the case.

Why it is worth recording rather than shrugging at

The general claim — add a hidden test layer one at a time, because a wholesale glob can bill the test layer for a non-test file — remains sound and should keep its place. What has changed is that the one concrete example it offers now demonstrates the opposite: the wholesale glob was adopted and cost nothing, because the errors were repaired instead of dodged.

This repo has been bitten twice today by exactly this shape — a hand-written passage that reads as current while the tree moved underneath it (#8024's stranded self-test fixture, and check-adr-merge-approval.mjs's header arguing for an account allowlist the ruling had superseded). Neither was a defect on the day it was written either.

Shapes a fix could take

1. Re-point the example at a package where the narrow glob is still the live choice. Requires finding one — worth measuring rather than assuming one exists.
2. Keep the example and mark it historical, naming [finding] examples/app-showcase/e2e/global-setup.ts is read by no tsc program — 6 errors, 3 of them holes in test/node-shim.d.ts #8062 / test(showcase): put e2e/global-setup.ts in a tsc program and fix what that reads #8178 as what changed and why the reasoning survived the change. Cheapest, and it preserves the concrete case, which is what makes the passage readable.
3. Generalise the passage and drop the named package entirely. Loses the concreteness that earns it its length.

⚠️ Whoever takes it: the six errors and the 3+3 split are recorded on #8062 and were re-measured on origin/main @ 7cfe7ad before the repair, so the historical claim is verifiable rather than folklore.

Establishment

Measured by the #8062 dev while repairing that card, and flagged to the PM rather than fixed because scripts/ was explicitly outside that card's file surface. Not independently re-read by the PM seat — the line number above is the dev's and should be confirmed against origin/main before editing.

Refs #8062, PR #8178, #7923, #7312.

[hotlong]

⚠️ 本卡的行号已被独立复核,并且发现漏了一处 —— 由 #7888 的 dev 在同一文件里作业时顺带测得(本席在派发令里要求他顺手确认,因为原行号来自另一位 dev、从未被证实)。

	结论
~1410	✅ 准确。过期示范位于 b5e09b21 的 1409–1412 行,在 measureTestDebt() 的 docblock 里
⚠️ 本卡漏掉的	同一说法还有第二份拷贝,在 :500,TEST_DEBT 前言中:「the wholesale glob would pull in e2e/global-setup.ts … and bill the test layer 6 errors that are not its own」

⇒ 两处一起改,或者都不改。 只修 1410 那处会留下一份措辞不同、断言相同的过期副本,那比原状更坏 —— 下一个读到 :500 的人没有任何线索知道它已被推翻。

顺带记一条本卡的方法论教训:立卡时我照抄了 dev 报的行号并标注「未经本席复核」,那是对的;但我没有问「这个说法在文件里出现了几次」。一份过期断言的第二份拷贝,和第一份一样会误导人,而且更难发现。凡是「某处散文已过期」的卡,立卡时就该带一次全文件计数。

⛔ 两处都未改动 —— #7888 是账本 diff,按其派发令第 5 条明确不搭车。本卡仍 open,finding,未指派。

---

Generated by Claude Code

[hotlong]

Finding-grading round: promoted to pm:queue as an S card, with direction 2 of the card's own list recommended — keep the worked example, mark it historical, naming #8062/PR #8178 as what changed and why the general claim (add hidden test layers one at a time) survived. Rationale: direction 1 requires hunting for a package where the narrow glob is still live (unmeasured, possibly empty set) and direction 3 trades away the concreteness that makes the passage worth its length; direction 2 is cheapest and the historical claim is verifiable (six errors re-measured on origin/main @ 7cfe7ad, recorded on #8062). Dev must confirm the ~line-1410 anchor against origin/main before editing — the line number is second-hand by the card's own admission. Type Task, domain:devx confirmed. Size/model suggestion: S, subagent, sonnet.

本评论来自分诊座位 Routine。

---

Generated by Claude Code