`zodShapeOf` 补完 union + `prefault` 后仍剩两格不对齐:不下探 `array` element / `record` valueType,且 depth 上限 12 是四个 walker 里最窄的

[os-zhuang]

观察类发现,来自 #6098 收尾时对四个同族 walker 的逐格对照。今天两格都被 fail-closed 兜住(解不出 shape 时 reachableVia 返回 'root-graph',宁可多要一个 tombstone),没有已知错误产物。记录在案,免得下一轮把「补齐了吗」当成已答之问。

#6098 补掉的是 union 分支与 prefault,这两格补完之后 zodShapeOf 与三个同族 walker 的包装器集合、lazy、pipe 方向、union 键面已经一致。剩下的不一致有两处,都不是 #6098 的范围。

格 1 —— 不下探 array element / record valueType

walker	array → element	record → valueType
scripts/lib/zod-graph.ts zodShapeOf	❌	❌
scripts/liveness/check-liveness.mts unwrap	❌	❌
src/kernel/metadata-authoring-lint.ts unwrap	❌	❌
src/system/metadata-form-zod-reconciliation.test.ts unwrap	✅ d.element	✅ d.valueType

所以这一格三比一,而且是唯一一个下探的那位跟另外三个不同——不能直接说「zodShapeOf 窄」。真正要先答的是语义问题:对 z.array(SomeObject) 这样的节点,「作者在这个节点上写的 shape」到底是不是元素的 shape?

• 对 keysOf 的问题(「这条路径下声明了哪些键」)答案显然是「是」;
• 对可达性的问题(「这个 def 的属性实例有没有出现在根闭包的桥里」)则等于把 array-of-X 的 def 视为携带 X 的 (name, instance) 对——这会新增桥项,是 #4001 战役的门测量 BFS 会误报「可达」:zod .describe() 共享 def 对象,任意单属性 bridge 把无关形状连起来 #5056 明确要求单独测量的那类改动。

现存标本:data/FilterArray(一个 lazy → array 的 emitted def,visited=false)今天 zodShapeOf 解不出 shape,落在 fail-closed 兜底上答 root-graph。#6098 实测的 1610 个 def 里只有它一个属于这一类。

格 2 —— depth 上限 12,是四个里最窄的

scripts/lib/zod-graph.ts                                12   (zodShapeOf 与 pipeInIsTransform 各一处)
scripts/liveness/check-liveness.mts                     16
src/kernel/metadata-authoring-lint.ts                   25
src/system/metadata-form-zod-reconciliation.test.ts     25

#6098 之后这一格的重要性上升了一点:union 分支是逐成员递归的(mergedUnionShape 对每个成员 depth + 1),嵌套 union 会比以前更快吃掉预算。今天没有截断——view 这条最深的链在 depth ≤ 3 就解完了——但「最窄的那个上限」正好落在喂删除门禁的这个 walker 上,方向不理想。

超限的后果同样是 return null → fail-closed,所以是保守而非错答。

与 #6098 的关系

#6098 的正文只点名了两格(union、prefault),这两格已随 PR #6222 合入并完成 #5056 测量轮。本条是那一轮收尾对照时发现的剩余项,单独记录而不是塞进那个 PR——因为格 1 会动桥表,必须有自己的测量轮。

建议

• 格 2(depth)可以独立成一个极小改动,但仍应带一次 gen:schema 前后 diff 才算数(理论上零移动,实测才能这么说)。
• 格 1 需要先回答语义问题,再决定是否对齐;如果对齐,必须走 #4001 战役的门测量 BFS 会误报「可达」:zod .describe() 共享 def 对象,任意单属性 bridge 把无关形状连起来 #5056 的完整测量协议(逐条解释新增桥项 + 逐条解释判定移动,两个方向都要看——fix(spec): zodShapeOf 补齐 union 分支与 prefault,并附 #5056 测量轮 (#6098) #6222 实测到的主要移动方向是 root-graph → null 这个放宽方向,不是 root-graph → derived-clone)。

参考:#6098 · PR #6222 · #5317 · #5056 · #6221

[os-project-manager]

Findings sweep (maintainer-authorized one-off, 2026-08-07 — registered on #6015): hold confirmed. Both remaining cells are fail-closed today (unresolved shapes err toward demanding tombstones), so the 3-vs-1 walker disagreement produces no wrong artifacts yet. Restart: a wrong product traced to the missing descent, or the next walker-family alignment card (pairs with #6221).

---

Generated by Claude Code

[os-zhuang]

Findings round — hold re-confirmed (finding + domain:spec-tooling unchanged). Stale-premise check on origin/main @ 424316b:

• Both cells are unchanged: zod-graph.ts still caps at depth > 12 (:109 and :224), check-liveness.mts still at depth > 16 (:239) — the narrowest limit still sits on the walker feeding the deletion gate; and zodShapeOf still does not descend array element / record valueType (the 3-vs-1 walker split stands).
• The safety argument stands: both cells fail closed (return null ⇒ root-graph ⇒ tombstone demanded), so the misalignment still produces conservative answers, not wrong ones; data/FilterArray remains the only def in the affected class.
• The open question is still semantic (is an array-of-X def's authored shape the element's shape — different answers for keysOf vs reachability), which is a measured-round change per #4001 战役的门测量 BFS 会误报「可达」:zod .describe() 共享 def 对象,任意单属性 bridge 把无关形状连起来 #5056's rule, not a drive-by.

Restart unchanged: a wrong product traced to the missing descent, or the next walker-family alignment card (pairs with #6221).

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Findings-triage re-check (quota round): HOLD maintained (finding stands, no label change).

Premise re-verified on origin/main @ 2c28df9: zodShapeOf live at packages/spec/scripts/lib/zod-graph.ts:223 (union-branch recursion at :201), and the four-walker family it is compared against still exists. The two remaining misaligned cells (no array element / record valueType descent; depth cap) are still fail-closed — an unresolvable shape returns 'root-graph', over-producing tombstone candidates rather than losing any. No known wrong artifact today. Held.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[claude]

Findings round (quota re-verification): HOLD maintained (finding + domain:spec-tooling unchanged). Re-checked @ origin/main a76a3af: both misaligned cells stand — zod-graph.ts still caps at depth > 12 (:109, :224) while check-liveness.mts allows 16 (line drift :239 → :274, now spelled if (!s || depth > 16)), and zodShapeOf still does not descend array element / record valueType (3-vs-1 walker split). Both cells fail closed (return null ⇒ root-graph ⇒ tombstone demanded), so answers remain conservative, not wrong. Restart unfired; pairs with #6221.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Findings cadence re-check @ origin/main 1530870: HOLD — finding stands.

Premise live: the depth cap is still 12 at both walker sites (packages/spec/scripts/lib/zod-graph.ts:109 and :224), and no array-element / record-valueType descent has been added to zodShapeOf (consumed unchanged from build-schemas.ts:1002/:1020). Measurement-tooling accuracy only; pairs with #6221 (same file) for any future promotion.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Findings cadence re-check @ origin/main 4ed4160: HOLD — finding stands, premise frozen.

Both cells re-verified unchanged: depth caps still 12/12 (zod-graph.ts:109, :224) vs 16 (check-liveness.mts:274) and 25/25 (metadata-authoring-lint.ts:90, reconciliation test :110); no descent into array element / record valueType in zodShapeOf (the :24 mention is the family docblock, not code); no commit has touched packages/spec/scripts/lib/zod-graph.ts since filing. Fail-closed protection intact — both cells still resolve to root-graph/null conservatism, no wrong artifact. Held because cell 1 needs its own #5056 measurement round (it moves the bridge table) and cell 2 alone is a micro-change the spec-tooling seat can fold into the next gen:schema-diffed pass; nothing user-facing waits on either.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Findings triage round, 2026-08-11 (maintainer-directed). Graded: hold — both remaining cells are fail-closed today (worst case: one extra tombstone demanded), and the array-element question is semantic, not mechanical — 3-vs-1 across the walker family with the ONE downward-walking implementation being the outlier. Restart: next walker-family card, or any measured wrong product from these cells.

---

Generated by Claude Code

[hotlong]

Findings-round disposition: hold carried (premise re-verified byte-exact: the 12/16/25/25 depth-cap spread across the four walkers stands, narrowest still on the gate-feeding walker; element/valueType descent still absent against a live zodShapeOf control; both cells still fail closed).

Pack recorded — PACK-zodgraph = {#6221, #6232} — see the #6221 comment of this round for the shared acceptance criterion. Restart condition: any commit touches packages/spec/scripts/lib/zod-graph.ts — promote the pack then.

本评论来自分诊座位（scheduled session session_01DLzE5XVRASxFotEU1h9cjH），不构成认领。

---

Generated by Claude Code

[hotlong]

Finding-grading round: closed, not planned (dissolving PACK-zodgraph = {#6221, #6232} together).

Both cells fail closed (worst case: one extra tombstone demanded), zero wrong artifacts ever measured, zero consumer pull; zod-graph.ts untouched since 2026-08-07 across ~90 merges, and the pack's own restart condition never fired. The 8 comments here are almost entirely re-verification passes returning the identical verdict — the card now costs more in triage than the defect ever did. Pack acceptance criterion, carried on close: zodShapeOf does not descend array element / record valueType, and the walker depth caps are 12/16/25/25. Whoever next edits zod-graph.ts: a 3-line docblock note (getter non-memoized; cap spread; both fail-closed) is where this knowledge should live — that rider serves the cards' stated purpose better than keeping them open.

---

Generated by Claude Code