fix(types): key the grid and report zod mirrors by their own declaration vocabulary (objectui#8516, objectui#8556)

.changeset/8505-grid-columns-breakpoint-narrowing.md

@@ -31,13 +31,17 @@ from a computed `Record<string, number>` still type-checks, and only fresh objec
 LITERALS meet excess-property checking. The narrowing is therefore a check on the
 authoring spelling, which is where the defect was authored.
 
-**The zod mirror is deliberately NOT narrowed here.** `zod/layout.zod.ts` still
-validates `columns` as `z.record(z.string(), z.number())`, so the JSON authoring
-face — `os-ui validate` / `check` in `@object-ui/cli`, the real consumer of these
-mirrors — still admits `{ xxl: 6 }`. That is reported for its own card, on a
-measurement: closing it ships runtime bytes into the console's `framework` chunk
-(`packages/(core|react|types)`), which measured 70,999 gzip bytes against its
-71,000 ceiling on this branch's base — one byte of headroom. The current reading is
-held visible by a handoff assertion in the new test, with a lit control next to it,
-so it flips to a refusal when that card lands rather than rotting into an
-assumption that both faces closed together.
+**The zod mirror WAS deliberately not narrowed here — corrected in #8573.** This
+paragraph said `zod/layout.zod.ts` still validates `columns` as
+`z.record(z.string(), z.number())`, so the JSON authoring face — `os-ui validate`
+/ `check` in `@object-ui/cli`, the real consumer of these mirrors — still admits
+`{ xxl: 6 }`. #8573 narrowed that mirror to a partial record over the six
+breakpoints (objectui#8516), so the JSON face now refuses `{ xxl: 6 }` exactly as
+the declaration above does. Its two supporting statements went with it, and #8573
+quotes each in its current state rather than re-arguing it: the byte measurement
+that deferred the narrowing — 70,999 gzip bytes against a 71,000 `framework`
+ceiling, one byte of headroom — is retired, because PR #8550 landed the
+maintainer's raise to `PER_CHUNK_GZIP_CEILINGS.framework` `100_000` against a
+`PER_CHUNK_BASELINE.framework` of `72_245`; and the handoff assertion described
+here as holding the current reading visible is flipped to a refusal in #8573, so
+it no longer asserts acceptance.

.changeset/8516-8556-mirror-partial-record-narrowing.md

@@ -0,0 +1,64 @@
+---
+'@object-ui/types': minor
+---
+
+**Breaking for authored metadata, at validation time:** two zod mirrors in
+`@object-ui/types/zod` now state their own declaration's key vocabulary instead
+of `string` (objectui#8516, objectui#8556).
+
+| key | mirror was | mirror is |
+| :-- | :--------- | :-------- |
+| `GridSchema.columns` (`zod/layout.zod.ts`) | `z.record(z.string(), z.number())` | a PARTIAL record over the six breakpoints |
+| `ReportComponentSchema.exportConfigs` (`zod/reports.zod.ts`) | `z.record(z.string(), ReportExportConfigSchema)` | a PARTIAL record over `ReportExportFormat` |
+
+Both declarations already stated the closed set — `columns` since objectui#8505,
+`exportConfigs` since objectui#6121 — so each mirror was accepting a spelling its
+own published type refuses. This is a pull-back to the declaration, not a new
+constraint: `os-ui validate` / `check` in `@object-ui/cli` are the real consumers
+of these mirrors, and they were passing documents `tsc` rejects.
+
+**What now fails that used to pass.** A `grid` whose responsive map is keyed
+outside `xs` `sm` `md` `lg` `xl` `2xl` — `{ xxl: 6 }` is the one to expect,
+because it is the Bootstrap/Ant spelling an author reaches for first — and a
+`report` whose `exportConfigs` is keyed outside `pdf` `excel` `csv` `json`
+`html`. Both were ALREADY broken at runtime: the grid renderer reads exactly the
+six keys and the export engine exactly the five formats, so a document the
+mirror used to accept rendered at the default column count, or exported nothing,
+with no error and no warning. The narrowing refuses documents that were already
+being silently ignored; it refuses nothing that works today.
+
+**Migration:** `xxl` is `2xl`; `XL` and `2XL` are `xl` and `2xl`. The refusal
+names the offending key — `Path: columns → xxl`, `Code: invalid_key` — through
+`@object-ui/cli`'s union-arm expansion.
+
+**Measured accept set in the corpus, before grading this.** Counted over every
+tracked file in this repository and in the `objectstack` sibling checkout, with
+test fixtures and changeset prose separated out rather than folded in:
+
+- grid nodes carrying an object-valued `columns` — **10 authored sites** (nine in
+  `content/docs`, `examples/schema-catalog` and `skills/objectui`, one in
+  `apps/site`), **all keyed inside the six breakpoints**. Every out-of-vocabulary
+  key in either tree sits in a test fixture written to document the defect, or in
+  objectui#8505's changeset quoting it.
+- `exportConfigs` — **one authored site**, `content/docs/core/report-schema.mdx`,
+  keyed `pdf` / `excel` / `csv`, all **in vocabulary**. None in `objectstack`.
+
+So this narrowing refuses **zero** documents that exist today, which is why it is
+graded `minor` with the break spelled out rather than escalated.
+
+**What it does NOT close.** These mirrors judge the document they are handed. A
+`grid` nested under another node's `children` still reaches
+`SchemaNodeSchema`, a lazy union over the passthrough base that does not re-enter
+the per-type arms, so `{ type: 'container', children: [{ type: 'grid', columns:
+{ xxl: 6 } }] }` still validates green. That is pre-existing and untouched here —
+named so this change is not read as closing the nested case.
+
+**The spelling is `z.partialRecord`, and that is load-bearing.** ⛔ Not
+`z.record(z.enum([…]), …)`: measured on zod 4.4.3, the plain record over an enum
+key REQUIRES every member, so `{ md: 2 }` stops parsing — it would trade this
+divergence for its exact opposite, and on `exportConfigs` it would re-impose the
+total-`Record` authoring face objectui#6121's maintainer ruling removed. That
+measurement is pinned executably in
+`__tests__/mirror-partial-record-narrowing-8516.test.ts`, at compile time (the
+inferred map is `Partial<Record<…>>`, not `Record<…>`) and at run time (one
+accepting row per member), so it cannot rot into folklore.

packages/types/src/__tests__/grid-columns-breakpoint-narrowing-8505.test.ts

@@ -252,28 +252,28 @@ describe('objectui#8505 — the narrowing broke no producer, and this is why', (
   });
 });
 
-/* ── (e) the other authoring face, REPORTED and left open by this card ────── */
+/* ── (e) the other authoring face — the handoff, now COLLECTED ────────────── */
 
-describe('objectui#8505 — the zod mirror still admits an open record, deliberately', () => {
-  it('the JSON face still accepts `{ xxl: 6 }` — the gap this card did not close', () => {
-    // NOT the desired end state. `GridSchema` in `zod/layout.zod.ts` is
-    // `z.record(z.string(), z.number())`, so `os-ui validate` / `check`
-    // (`@object-ui/cli`, the real consumer of these mirrors) still passes a
-    // grid document whose map is keyed by a spelling no renderer reads.
+describe('objectui#8505 — the zod mirror now refuses what the declaration refuses', () => {
+  it('the JSON face refuses `{ xxl: 6 }` — the gap objectui#8516 closed', () => {
+    // THE HANDOFF PIN, FLIPPED (objectui#8516, the objectui#7070 convention).
+    // It asserted `success: true` for the whole of objectui#8505's life, with
+    // the reason stated rather than assumed: that card narrowed the TypeScript
+    // face only, and this block held the mirror's remaining reading visible so
+    // it could not rot into a silent assumption that both faces closed
+    // together. objectui#8516 narrowed the mirror, so the block is collected
+    // here rather than deleted — the flip IS the record that the handoff was
+    // honoured, and a deletion would leave nothing that fails if the mirror is
+    // ever widened back.
     //
-    // Left open on a measurement, not a preference: closing it ships runtime
-    // bytes into the console's `framework` chunk (`packages/(core|react|types)`),
-    // which measured 70,999 gzip bytes against its 71,000 ceiling on this
-    // branch's base — one byte of headroom. A zod narrowing here would turn
-    // `Bundle Analysis` red and make its own fix a byte-hunt through two other
-    // packages, which is a different card. The TypeScript narrowing this card
-    // ships costs zero runtime bytes.
-    //
-    // This assertion is the HANDOFF (the objectui#7070 convention): when that
-    // card lands, this block flips to a refusal rather than quietly agreeing
-    // with whatever the mirror ends up doing.
+    // ⚠️ The blocker this block used to cite is GONE, and the obsolete number
+    // is left named on purpose: it said a narrowing was unaffordable because
+    // the console's `framework` chunk measured 70,999 gzip bytes against a
+    // 71,000 ceiling. PR #8550 landed the maintainer's raise — that ceiling is
+    // 100,000 against a re-derived 72,245 baseline — so the narrowing cost no
+    // byte-hunt at all.
     const r = GridZodMirror.safeParse({ type: 'grid', columns: { xxl: 6 } });
-    expect(r.success).toBe(true);
+    expect(r.success).toBe(false);
   });
 
   it('CONTROL — the mirror is live: it refuses a non-numeric column count', () => {
@@ -282,4 +282,15 @@ describe('objectui#8505 — the zod mirror still admits an open record, delibera
     const r = GridZodMirror.safeParse({ type: 'grid', columns: { xs: 'two' } });
     expect(r.success).toBe(false);
   });
+
+  it('CONTROL — and it still ACCEPTS a partial breakpoint map', () => {
+    // ⚠️ Added with the flip, and load-bearing because of it. Both readings
+    // above are now `false`, so on their own they are equally green against a
+    // mirror that refuses EVERYTHING — the caricature the header's table warns
+    // about, one authoring face down. This is also the overshoot guard: the
+    // `z.record(z.enum([…]), …)` spelling objectui#8516 refused reddens exactly
+    // here, because zod 4 makes it require all six members.
+    const r = GridZodMirror.safeParse({ type: 'grid', columns: { md: 2 } });
+    expect(r.success).toBe(true);
+  });
 });