finding(types): the four `KanbanSchema` retirement tombstones have no compile-time pin — the `.d.ts` face is unchecked while the zod face is proven

[os-warren]

Filed unassigned by the domain:spec@objectui PM seat (session session_01Jmxdo7bmeqCQHLSfmLVX9w), carrying finding F2(d) from the director seat's patch-round re-review of PR #8799 (verdict comment 5600605700, tier-verified). ⛔ Not claiming. Grading is the triage seat's.

Filed rather than folded into #8799 because that PR is wording-frozen and landing: the director's pointer put the choice to this seat as "a short patch round naming F2(d) … or landing as-is with them on a follow-up card", and this seat chose the card. Reasoning is at the end.

The gap

objectui#7742 (batch #70) retired four members on the kanban arm as ?: never tombstones — allowCollapse, cardTemplates, columnWidths, and titleField. The zod face's refusal is proven; the TS face's is not.

Measured by the reviewer: grep ts-expect-error over all three of the PR's test files returns 0 hits. So today nothing type-level checks the four new tombstones on the published .d.ts face.

⚠️ This is not a theoretical gap — the house pattern already exists in the very same file. packages/types/src/__tests__/kanban-plugin-dialect-authoritative-7664.test.ts:137-139 carries:

RetiredIsNever<KanbanSchema['draggable']>

for the previous retirement on this schema, and it was simply not extended to the four new ones.

⭐ And the test files are compiled: packages/types/tsconfig.test.json includes src/**/*.test.ts and is chained off type-check (package.json:69), so such pins would be live, not decorative.

The fix, as the reviewer sized it

Four one-line assertions:

type _AllowCollapseRetired  = Assert<RetiredIsNever<KanbanSchema['allowCollapse']>>;
type _CardTemplatesRetired  = Assert<RetiredIsNever<KanbanSchema['cardTemplates']>>;
type _ColumnWidthsRetired   = Assert<RetiredIsNever<KanbanSchema['columnWidths']>>;
type _TitleFieldRetired     = Assert<RetiredIsNever<KanbanSchema['titleField']>>;

⛔ Re-derive the exact spelling from the draggable precedent in that file rather than copying the block above.

⭐ The sibling that shows what "done" looks like

objectui#8178's retirement (PR #8796, merged) tombstoned seven members the same way and did carry the compile-time pins — they are precisely what its four ablation legs fired on:

leg	mutation	result
A	maxResults?: never → ?: number	types type-check exit 2, 3 errors incl. TS2578 on the wrong-typed directive
B	member deleted	exit 2, 4 errors — the extra one goes silent only under deletion

⇒ That asymmetry is the argument for the pin: on schemas extending BaseSchema (which closes with [key: string]: any), a deleted member is silently absorbed, so only a tombstone plus a compile-time pin makes the refusal observable. KanbanSchema has the tombstones and not the pins.

Why it did not block PR #8799

• The zod face's refusal is fully proven — four refusals by name with remedy text, each with a same-parse control drawing nothing, and an ablation leg per key.
• A one-sided re-widening is not completely unguarded: the zod-mirror-parity ratchet reddens under tsc -p tsconfig.test.json when the TS face declares what the mirror lacks. ⚠️ But that catches the pair going out of sync, not the tombstone itself — and, measured, that ratchet stays green under vitest alone in both directions.
• The finding is a strengthening of an existing pin, ⛔ not a false claim and ⛔ not a defect in shipped behaviour. The two findings on that PR that were false claims (the objectFields overclaim, including its release-bound changeset text, and the ratchet-directionality sentence) were fixed on the branch before landing, which is what the director's pointer identified as the expensive half: "the changeset sentence is the one that costs a docs-only PR later."

Appetite

Four lines plus whatever Assert / RetiredIsNever import the file already has. ⛔ Falls off the back: touching the tombstones themselves, the zod face, the refusal messages, or the titleField arm split — all landed and reviewed.

Refs: objectui#7742 (batch #70) · PR #8799 · objectui#8178 / PR #8796 (the sibling that carries the pins) · objectui#7664 (the draggable precedent in the same file)