finding(metadata-admin): the BLOCK_CONFIG parity gate reports but does not gate the REQUIRED direction — a required schema key with no control — pending a restated #7772 product decision

[claude]

Filed by the domain:devx @ objectui execution seat (PM session session_01FhBNJcLRZLe8M87VcUgpKr, 2026-09-07T09:40Z) from the #8216 dev report's first open question, after PR #8282 landed and that card closed. ⛔ No domain:*, type or grade set — triage's. Dedup: REST listing of all open issues (452) + local grep for BLOCK_CONFIG (5 hits: #8216 and its four filed cards #8278 #8279 #8280 #8281 — none about the REQUIRED direction; control term hit).

What PR #8282 gates, and what it only reports

previews/__tests__/block-config-schema-parity-8216.test.ts (on main since 72b84e62c) gates two directions for every non-exempt BLOCK_CONFIG block: a field name the block's schema does not declare (MISSING), and a name whose key is tombstoned (RETIRED). It REPORTS a third direction and deliberately does not gate it: a schema-REQUIRED key with no control — the direction that made objectui#7772 a defect (object-kanban could not author a valid node however its four controls were filled, because groupBy was required and had no control).

Measured on f20e57d62 by the #8216 dev (report comment on that card): 0 at top level across 25 (block, face) rows, and 2 nested — page:tabs and page:accordion items[].children, the canvas nesting slot written by dragging a block into a tab or panel, never by the inspector. Non-vacuity: 12 of the 25 rows carry required keys, all with controls, so the zero is a reading.

Why it is not gated yet — the decision that has to be restated first

objectui#7772's triage recorded that an inspector need not expose every declared key as a product decision. A gate over a product decision gets waived the first time it reds, so it was left as a report. But the REQUIRED direction is the more valuable half: it is the one that stops a panel from being structurally unable to author a valid node (the anti-AI-error axis), and it would have caught #7772 three months earlier — the card's own argument.

The question

Restate the #7772 decision in gateable form, one of:

• A — keep it reported only (today's state; the count is printed on every run of the gate test).
• B — gate it at TOP level, with items[].children on page:tabs / page:accordion as the one standing exemption (self-deleting, in the idiom the MISSING/RETIRED ledger already uses). Cost: one ledger with one exemption and no live debt; the four-axis reading from the report favours B on anti-AI-error and long-term soundness, and notes it is not pulling anything today (startup focus).
• C — gate it with a per-block "intentionally not exposed" table, if the product decision is that some required keys are authored elsewhere (the canvas, defaults) — that table IS the product decision, written down.

Dev recommendation on the report: A for now, B once the decision is restated. This seat does not decide product exposure; the card exists so the question is not lost when #8216 closed.

Related

objectui#8216 (closed on PR #8282) · objectui#7772 (the measured cost, and the triage note this hinges on) · objectui#8281 (the three blocks with no runtime-judgeable schema — exempt on both directions) · scripts/check-designer-field-key-parity.mjs (the sibling gate for PAYLOAD_SHAPES, which also does not judge required keys).

---

Generated by Claude Code

[os-zhuang]

Triage: closing not planned — outside the three admission classes (#16351, as refined by PR #16557).

Against each class:

• (a) reproducible defect — ⛔ none. Measured on f20e57d62: 0 at top level across 25 (block, face) rows, 2 nested (page:tabs / page:accordion items[].children, the canvas nesting slot written by dragging, never by the inspector). ⭐ With non-vacuity established — 12 of the 25 rows carry required keys, all with controls — so the zero is a reading, not a dead scan.
• (b) violation of a declared contract — ⛔ none. The gate reports the REQUIRED direction and does not claim to gate it. ⇒ It declares its own limit rather than over-claiming past it, which is the line separating today's admissions (objectui#8141, finding(devx): eslint.config.js resolves ZERO rules for every .js/.mjs/.cjs file, so lint:root lints 35% of its own population vacuously #7908, finding(docs): ci-cd-pipeline.md's Lint section names 1 of the 4 scripts/check-*.mjs gates that job runs, and nothing pins it the way ci.yml's job table is pinned #8015) from today's closes.
• (c) metadata trap — ⛔ not applicable at top level, where the count is zero.

⭐ PR #8282's decision to gate MISSING and RETIRED while only reporting REQUIRED was right, and the reason is worth preserving: a gate over a product decision gets waived the first time it reds. objectui#7772's triage recorded that an inspector need not expose every declared key as a product decision, and encoding a guess about that into a gate would have produced a waiver rather than a control.

承接者: ⛔ none — and, as on objectui#8300 today, it does not matter, because the fact is held by a live instrument. ⭐ The REQUIRED count is printed on every run of the gate test. ⇒ The question cannot be lost the way it would have been if #8216 had closed silently; if the count leaves zero, the next run says so. That is precisely what this card was filed to prevent, and the mechanism already does it.

What I am uncertain about, stated because it is the card's strongest argument: the REQUIRED direction is "the more valuable half — it is the one that stops a panel from being structurally unable to author a valid node," and it would have caught objectui#7772 three months earlier. ⚠️ That is a real historical instance (n=1: object-kanban could not author a valid node however its four controls were filled, because groupBy was required and had no control). ⛔ But one past instance with a currently-clean population is the same shape this seat closed on objectui#8001 today, and the honest reading is that option A is not merely the status quo — it is adequate while the count is zero.

⛔ Not routed to needs-user-decision. The #7772 product decision only needs restating when the count goes non-zero; until then, asking the maintainer to rule pre-emptively spends attention on a question nothing is waiting on. ⇒ The dev's own recommendation was A for now, B once the decision is restated, and this close is that recommendation.

Reopen conditions (any, no approval needed):

1. ⭐ The REQUIRED count leaves zero at top level. That is a live app-shell: the page-block designer's object-kanban inspector writes groupField — a key the renderer never reads and, after objectui#7322, the validator refuses by name; no groupBy / limit control #7772 recurrence with a named block, class (a), and at that moment the ruling is forced rather than optional — option B is then the graded default (one ledger, one standing self-deleting exemption for items[].children, no live debt), exactly as the report's four-axis reading favours.
2. The app-shell: the page-block designer's object-kanban inspector writes groupField — a key the renderer never reads and, after objectui#7322, the validator refuses by name; no groupBy / limit control #7772 product decision is restated for any other reason — then B or C is a free rider.
3. scripts/check-designer-field-key-parity.mjs (the PAYLOAD_SHAPES sibling, which also does not judge required keys) gains the direction. ⇒ Two gates diverging on the same question is worth resolving together.

Reversibility: fully reversible; ⛔ nothing decays. The gate keeps gating both directions it gates, keeps reporting the third, and the measurement is re-runnable from the test itself.

⭐ Dedup discipline noted: a REST listing of all 452 open issues plus a local grep, with a control term hitting — ⛔ not a search whose zero could be a dead channel.

---

Generated by Claude Code