diff --git a/.changeset/8801-object-kanban-allow-collapse-retired.md b/.changeset/8801-object-kanban-allow-collapse-retired.md
index f897016a3f..39413d0b60 100644
--- a/.changeset/8801-object-kanban-allow-collapse-retired.md
+++ b/.changeset/8801-object-kanban-allow-collapse-retired.md
@@ -59,17 +59,42 @@ SWIMLANE collapse remains the viewer's alone: `KanbanImpl` collapses a swimlane
 row when its header button is clicked and persists that set per `swimlaneField`,
 and no authored key reaches it.
 
-**Migration, in this repository: no authored document changed.** Over every
-tracked file, the only occurrences of the name were the two declarations retired
-here, one pin, one docs row and two historical changesets — no board, no
-example, no fixture authors it, and `@object-ui/plugin-kanban` names it in ZERO
-files against firing controls in the same pass (`groupBy` 45 files,
-`conditionalFormatting` 8, `quickAdd` 6, `coverImageField` 3). The pin that
-moved is `bare-kanban-node-key-retired-8802.test.ts`, whose suite 3 asserted
-this arm still ACCEPTED the key; that row is now a refusal assertion, at its
-site, with the reason stated — its three sibling rows stay accepting, because
-the claim that suite makes (batch #70's refusals were arm-scoped and cannot
-cross an arm) is still true and this retirement is not that claim.
+**Migration, in this repository: no authored document changed.** No board, no
+example and no fixture authors the key. Scanned 2026-09-17 over every tracked
+file — with a whitespace-tolerant probe rather than a line-anchored grep, since
+the name wraps across lines in prose — every occurrence is the retirement
+talking about itself, and they are these:
+
+- the declarations retired here — `packages/types/src/objectql.ts` and its
+  mirror `packages/types/src/zod/objectql.zod.ts`;
+- the pins that assert the retirement —
+  `object-kanban-allow-collapse-retired-8801.test.ts` and
+  `bare-kanban-node-key-retired-8802.test.ts`;
+- a comment in `packages/types/src/zod/complex.zod.ts`, recording that the
+  deleted `retiredZeroReadKanbanKey` helper once carried this spelling on the
+  SIBLING arm;
+- one row of `content/docs/api/schema-reference.md`;
+- the `.changeset/` release notes that discuss it — this one, the two
+  historical entries covering the sibling arm's own spelling, and objectui#9629's
+  note recording the correction to this paragraph.
+
+`@object-ui/plugin-kanban` names the key in ZERO files, against live sibling
+keys (`groupBy`, `conditionalFormatting`, `quickAdd`, `coverImageField`) firing
+as controls on the same walk. ⛔ Their counts are deliberately not written here:
+this file publishes VERBATIM into the CHANGELOG at an unknown future date, and a
+figure frozen there is derived once and re-derived never. The walk is the
+instrument and it re-runs on every test run — the `it` named
+"`@object-ui/plugin-kanban` names it in ZERO files, with controls firing in the
+same pass", in `object-kanban-allow-collapse-retired-8801.test.ts`.
+
+The pin that moved is `bare-kanban-node-key-retired-8802.test.ts`, whose suite 3
+asserted this arm still ACCEPTED the key. That stale row is gone, and it was not
+replaced by a refusal row in the same place: the refusal belongs to this key's
+own pin, which asserts it with the message, the `invalid_type` code and its own
+firing controls, so the claim is pinned once rather than in two files. That
+suite's accepting rows stay, because the claim it makes (batch #70's refusals
+were arm-scoped and cannot cross an arm) is still true and this retirement is
+not that claim.
 
 ⚠️ This repository's census cannot see a TypeScript consumer outside it that
 wrote the key. Such a consumer gets a compile error naming the member, which is
diff --git a/.changeset/9629-9621-prose-residuals.md b/.changeset/9629-9621-prose-residuals.md
new file mode 100644
index 0000000000..d1fec8c299
--- /dev/null
+++ b/.changeset/9629-9621-prose-residuals.md
@@ -0,0 +1,39 @@
+---
+---
+
+Internal only — no published package moves, so this declares no release
+(objectui#9629). Three prose corrections to what objectui#9621 left behind, each
+named as NON-BLOCKING by that PR's isolated at-tier contract review (VERDICT
+PASS). ⛔ None of them is a contract defect, no published face moves in either
+direction, and no assertion of either retirement pin was weakened.
+
+1. **A docblock stopped writing an instrument's answer down.** The
+   `ObjectKanbanSchema.allowCollapse` tombstone's docblock recorded per-key
+   control counts as text. It now names the walk that derives them — the `it`
+   called "`@object-ui/plugin-kanban` names it in ZERO files, with controls
+   firing in the same pass", in
+   `packages/types/src/__tests__/object-kanban-allow-collapse-retired-8801.test.ts`
+   — and names the control keys without their counts, per commandment #9. The
+   figures it carried had already drifted before they landed, which is the case
+   that rule calls the dangerous one: a reader who spot-checks a still-plausible
+   number confirms it and is still wrong.
+
+2. **A suite title stopped contradicting its own contents.** In
+   `packages/types/src/__tests__/bare-kanban-node-key-retired-8802.test.ts`,
+   suite 3 was titled as though the sibling `object-kanban` arm kept every
+   verdict it ever had, while carrying a row asserting that one of them had
+   changed. The title now states the claim the suite actually measures: retiring
+   the bare `kanban` arm moved no verdict on the sibling arm. The restating
+   refusal row is gone rather than re-titled — it duplicated, in another card's
+   file, a refusal that `object-kanban-allow-collapse-retired-8801.test.ts`
+   already asserts together with its message, its `invalid_type` code and its
+   own firing controls. A comment in its place records why it is absent, so it
+   is not restored in either direction.
+
+3. **A changeset enumeration stopped publishing short.** The objectui#8801
+   release note listed where the retired name still occurs and omitted a comment
+   in `packages/types/src/zod/complex.zod.ts`. Re-scanned with a
+   whitespace-tolerant probe — a line-anchored grep cannot see prose that wraps
+   mid-name — the paragraph now names each site rather than counting them, and
+   carries the date it was taken, because that body publishes verbatim into the
+   CHANGELOG at an unknown future date. Its frontmatter is untouched.
diff --git a/packages/types/src/__tests__/bare-kanban-node-key-retired-8802.test.ts b/packages/types/src/__tests__/bare-kanban-node-key-retired-8802.test.ts
index 421868b3a8..70474b9889 100644
--- a/packages/types/src/__tests__/bare-kanban-node-key-retired-8802.test.ts
+++ b/packages/types/src/__tests__/bare-kanban-node-key-retired-8802.test.ts
@@ -27,8 +27,11 @@
  *     it, and a 20-member census of it.
  *
  * ⭐ The one claim inside them that was NOT about the retired arm is carried
- * forward here as suite 3: the SIBLING `object-kanban` arm's verdicts must not
- * move. That is what turns this from a deletion into a retirement.
+ * forward here as suite 3: retiring this arm must not move the SIBLING
+ * `object-kanban` arm's verdicts. That is what turns this from a deletion into
+ * a retirement. ⛔ It is not the wider claim that those verdicts never move at
+ * all — `allowCollapse` later moved on that arm's own ruling (objectui#8801),
+ * which suite 3's comments cover.
  *
  * ## ⚠️ The mechanism, which is NOT the passthrough rule
  *
@@ -153,7 +156,7 @@ describe('suite 2 — the refusal is not a validator that turned strict', () =>
   });
 });
 
-describe('suite 3 — the SIBLING `object-kanban` arm keeps every verdict it had', () => {
+describe('suite 3 — retiring the bare `kanban` arm moved NO verdict on the SIBLING `object-kanban` arm', () => {
   // Carried forward from `kanban-arm-batch70-7742.test.ts`, whose `titleField`
   // pair is the reason this suite exists: the batch #70 refusals were
   // ARM-SCOPED, so retiring the arm must not move the sibling's answers.
@@ -165,24 +168,22 @@ describe('suite 3 — the SIBLING `object-kanban` arm keeps every verdict it had
     expect(refusals({ type: 'object-kanban', objectName: 'tasks', groupBy: 'status', [key]: value })).toEqual([]);
   });
 
-  // ⭐ `allowCollapse` LEFT this row, and the distinction is the whole point of
-  // the suite rather than an exception to it. Its verdict did not move because
-  // the `kanban` arm retired — that is precisely what this suite denies, and it
-  // went on being accepted here for a week after objectui#8802 landed. It moved
-  // because objectui#8801 ruled on THIS arm on its own protocol reading
+  // ⭐ `allowCollapse` is NOT a row in this suite, and its absence is the point
+  // rather than an omission. It sat among the accepting rows above until
+  // objectui#8801 ruled on THIS arm's own protocol reading
   // (`ComponentPropsMap['object-kanban']` never declared the key; no registered
-  // board reads it). ⛔ Do not read this line as batch #70 finally reaching the
-  // sibling: an arm-scoped refusal still cannot cross an arm, and the three rows
-  // above are still the measurement that says so.
-  it('REFUSES `allowCollapse` — not inherited from the retired arm, ruled for this one (objectui#8801)', () => {
-    const found = refusals({
-      type: 'object-kanban',
-      objectName: 'tasks',
-      groupBy: 'status',
-      allowCollapse: true,
-    });
-    expect(found.filter((f) => f.path === 'allowCollapse')).not.toEqual([]);
-  });
+  // board reads it) and the key became a refusal. ⛔ Do not read that as batch
+  // #70 finally reaching the sibling: its verdict did NOT move because the bare
+  // `kanban` arm retired — an arm-scoped refusal still cannot cross an arm, and
+  // the three rows above are the measurement that says so.
+  //
+  // ⛔ Do not restore it here in either direction. The refusal — with its
+  // message, its `invalid_type` code, its `.describe()` channel and its own
+  // firing controls — is asserted by
+  // `object-kanban-allow-collapse-retired-8801.test.ts`, in the `it` named
+  // "refuses the value %p at the key's OWN path, with the prescription".
+  // Restating it here would put one pin in two files, which is the shape that
+  // let the stale ACCEPTING row survive here for a week after objectui#8802.
 
   it('and still REFUSES what it always refused — `groupField`, its own tombstone', () => {
     // The other half: suite 3 would be vacuous if the sibling arm accepted
diff --git a/packages/types/src/objectql.ts b/packages/types/src/objectql.ts
index cd6a280fde..22673c6f69 100644
--- a/packages/types/src/objectql.ts
+++ b/packages/types/src/objectql.ts
@@ -3820,12 +3820,26 @@ export interface ObjectKanbanSchema extends BaseSchema {
    *
    * ## Zero read sites, and both unnamed channels terminate
    *
-   * `@object-ui/plugin-kanban` names the token in ZERO files — the firing
-   * controls on the same instrument and the same run are `groupBy` (45 files),
-   * `conditionalFormatting` (8), `quickAdd` (6) and `coverImageField` (3), so
-   * the zero is a reading and not a dead grep. A source grep alone cannot
-   * answer the inertness question, because a renderer may consume a key it
-   * never names; both such channels were traced to their ends:
+   * `@object-ui/plugin-kanban` names the token in ZERO files, and that zero is
+   * a reading rather than a dead grep because live sibling keys fire as
+   * controls on the same instrument and the same run — `groupBy`,
+   * `conditionalFormatting`, `quickAdd` and `coverImageField`.
+   *
+   * ⛔ Their per-key counts are deliberately NOT written here (AGENTS.md #9).
+   * A figure in prose is derived once and re-derived never: the four that used
+   * to stand in this paragraph had already drifted by the time they landed, and
+   * a reader who spot-checked them would have confirmed a number the population
+   * underneath had already moved out from under. Point at the instrument
+   * instead — the walk re-runs on every test run, as the `it` named
+   * "`@object-ui/plugin-kanban` names it in ZERO files, with controls firing in
+   * the same pass" in
+   * `packages/types/src/__tests__/object-kanban-allow-collapse-retired-8801.test.ts`.
+   * The ZERO above is that test's own assertion rather than a remembered
+   * reading, which is why it stays here while the control counts do not.
+   *
+   * A source grep alone cannot answer the inertness question, because a
+   * renderer may consume a key it never names; both such channels were traced
+   * to their ends:
    *
    *   - the PROP channel — `SchemaRenderer` spreads every non-metadata key as a
    *     React prop, `ObjectKanbanRenderer` destructures `schema` and forwards