[rorychatt]

What this records

The independent check that closed PR 36 (9047807, tag v0.2.1-r2) left a set of minor findings standing. This PR writes them down where a maintainer will find them. Docs only: docs/ISSUES.md (D9, D14 and two new groups D14a to D14f and D15a to D15c) and one paragraph of docs/brand/WORKFLOWS-LISTS-DASHBOARDS.md. No product code, test or script is changed. Nothing is fixed here; every limit says why it was left and what the fix would be.

Where each finding landed

Finding	Landed in	Note
F-A (callback parameter makes a translator)	docs/ISSUES.md D14a, pointer in the D14 row	The reader rated it medium, both skeptics low. Effect, proposed fix and negative fixtures written; follow-up, not done
F-B (wrappers, parametersOf)	D14b	Also records one more shape from the same check, export const title = (t("x") + "y"), which errs the other way (a missed translator); it is the same function
F-C (unspaced <)	D14c
F-D (scopeAt reads non-parameter groups)	D14d
F-E (unpinned guard code, vacuous row)	D14e	Six mutants, each with the input that kills it. The finding's sixth mutant ("comma cut in marked()") was dropped: marked() has no comma cut, so there is nothing to pin; the k = first - 1 mutant of the mutation matrix (M20) took its place
F-F (block comment with code between)	D14f (1)
BD4-4 (161-column header line)	D14f (2)
BD4-1 (D9 list leaves out the two-factor panel)	D9	Added packages/ui/src/twofactor.tsx, ImportDialog.tsx, connectors/ConnectionDetail.tsx
BD4-2 (GROUPABLE sentence)	WORKFLOWS-LISTS-DASHBOARDS.md	Reworded: GROUPABLE limits only what is picked; the Kanban totals and useCurrencyOf also group by currency
DT-1 (test 16 ends too early)	D15a	Proposed waitForResponse fix included. The 404 teardown race is written as a possibility, not a defect
M26 (useRecord glue untested)	D15b	Proposed test: extract the glue into a pure function and test it
M20 (marker above multi-line block)	D14e (6)
M14a/b (PlanDialog wording)	D9 sentence + D15c	Proposed e2e test under a non-English language
BD4-3 (PR 36 description)	not in this PR	It is the text of the PR 36 body, not a repo file

Three things I added that were not in the list, so you can drop them: DriveBrowser.tsx in the D9 list (it came in with PR 39, after the PR 36 check, and the D9 search finds it); a correction of D9's own wording about connectors/Sections.tsx (its checklist keeps its sentence in setState({ error }), which the setError(/setFailure( search does not find; only the register export is found); and the (t("x") + "y") shape in D14b.

What I verified, and how

Every sentence was checked against origin/main at 9555614 before it was written.

• The guard (i18nGuard.ts): ran it from a scratch script outside the repo (Node 26 type stripping) on fixtures built like the ones in i18nDeps.test.ts, and on the real tree (132 files, 60 hooks, 0 flagged). Reproduced: F-A for six parameter shapes ((t) =>, t =>, ({ t }) =>, function (t), catch (t), for (const t of ...)), plus the real-tree effect (Tabs, Navigation in the translator set; dismiss, notify, submitQuick among the locals; notify a member; a scratch useMemo over <Tabs> and over d.notify flagged); F-B for the five wrapper shapes and the parenthesised initialiser; F-C for < and <= unspaced, with the spaced and &&/> forms still flagged; F-D for all eight forms; F-F.
• The test file (i18nDeps.test.ts): ran it through a small stand-in for vite-plus/test: 30 tests, 139 assertions, 0 failed. Then six mutants of the guard (declared(header) dropped, the t => branch and the function branch of parametersOf dropped, <= 1 to <= 2, the => guard of argumentComplete dropped, k = first - 1 to k--): all six still pass 139/139, and each is killed by the input named in D14e. A sanity mutant (<= 0) is killed, so the stand-in does fail when it should. The fourth header of the F1 loop flags once under the old reading of the header too; with the body fmt.money(1) it flags once now and zero times under the old reading.
• Greps for the rest: line and function names in twofactor.tsx, Security.tsx, ImportDialog.tsx, ConnectionDetail.tsx, Sections.tsx, PlanDialog.tsx, useServer.ts, lastRecord.ts, KanbanView.tsx, useAggregate.ts, aggregate.rs (cmp_keys, TextOrder), F::new (default FieldKind::Text) and the sales and invoicing currency fields; the e2e fixtures (allowHttpErrors, the module.uninstall teardown) and UnknownCommand mapping to 404; no test file under web/ mentions PlanDialog or setFailure; no DOM or hook test library in the back office.
• Collation: the 162 ISO 4217 codes of Intl.supportedValuesOf("currency") sort into one order in all seven languages with Intl.Collator(lang, { numeric: true, sensitivity: "base" }). That is the browser collator; the server's icu_collator was not run.
• Merge safety: all edits are in the D section of ISSUES.md and one paragraph of the brand doc. git merge-tree against feat/oidc-core (PR 37), feat/mileage-ios and feat/mileage-app auto-merges docs/ISSUES.md without conflict (those branches have conflicts of their own elsewhere, none in these two files).
• Markdown: table rows of the new groups have the same number of columns as their headers (checked with a script). Four older rows (B3, B5, I1, K4) have unescaped pipes inside code spans; I did not touch them.

Not run, or not reproduced

• Not run: vp check, vp test, the end-to-end suite, CI, cargo, a browser. The markdown was not rendered.
• Taken from the check, not repeated: that the M26 change and the M14a/b changes pass the end-to-end and unit suites (D15b, D15c say so). The 404 race of D15a is a possibility read from the code; neither skeptic reproduced it.
• D15b: that the M26 change is visible only after a longer sequence is read from the code, not run. The sequence is written out in the row.
• Nothing from the list failed to reproduce. One mutant and one proposed fixture of F-E were wrong as written (see the table and D14e).

🤖 Generated with Claude Code