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 #8306

Activity

1. claude

   added

   finding

   on Sep 7, 2026
2. os-zhuang commented on Sep 7, 2026

   

   os-zhuang

   on Sep 7, 2026 – with Claude

   Contributor

   More actions

   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
3. os-zhuang

   closed this as not plannedon Sep 7, 2026