[Decision] Declare `renderCellEditor` and schema-level `cellClassName` on `DataTableSchema`? — the two live undeclared keys the #6459 census measured

[os-sam]

Filed by the domain:ui execution seat (PM session session_013hfmP9hoMd3dJwTh85J4yB, R28), executing the 2026-08-29 ruling's own closing clause on #6459:

Any contract gap the census reveals that needs a @object-ui/types decision (a key data-table really reads but DataTableSchema doesn't declare, or vice versa) comes back to the decision inbox as its own small card rather than being decided in-lane — the #6424 family precedent.

⛔ This seat is not deciding it. packages/types is a published surface, this widens the accept set, so it is clause-② and the human floor.

The measurement (#6459 census, on main a04d7c65, AST pass)

Writers pass 47 distinct schema-level keys into dataTableSchema; DataTableSchema declares 73 named members over its chain, excluding the [key: string]: any at base.ts:382. The difference in the "passed but not declared" direction is exactly two, both with live readers:

key	live reader	consequence of staying undeclared
renderCellEditor	data-table.tsx, called through its own (schema as any) cast to hand cell editing to the returned widget	cell editing falls back to the built-in text/number/date inputs
schema-level cellClassName	data-table.tsx, destructured and folded into every body cell's className	the grid's row-density styling never reaches the cells

⇒ both are in-use capabilities that are simply not declared. Not speculative keys.

⚠️ Two corrections to the 2026-08-29 ruling's own text, forced by this census:

• its step-1 suspect list is stale: four of the five keys it named — borderless (:657), disableInnerScroll (:666), selectionResetKey (:905), singleClickEdit (:921) — are declared today, leaving only renderCellEditor;
• and it never named schema-level cellClassName, which the census adds.

⭐ Stability check, so this is not a snapshot: the same instrument re-run at PR #6637's merge commit f241a4df gives a byte-identical direction-A set (keys 47→47, declared 72→73, the sole addition being bind from #6574). The set has not moved.

Options

	what	cost
A	Declare both keys on DataTableSchema + document them; data-table.tsx's (schema as any) cast at :2269 can then be dropped	widens a published type face ⇒ clause-②, contract-review tier; one types change + docs
B	Keep the seam-level hold introduced by PR #6637 — the two keys stay as ObjectGridDataTableSchemaHolds inside plugin-grid	zero cost, but a published renderer keeps reading undeclared keys through a cast, and authors get no completion or checking for either key

四维分析

① 实际业务需求。 两个键今天都在生产路径上被读,且各自缺失都有可感知后果:renderCellEditor 缺失 ⇒ 单元格编辑退回内置输入;cellClassName 缺失 ⇒ 行高密度样式到不了单元格。⇒ A 不是新增能力,是把既有行为说成真话。⚠️ 置信缺口如实标注:仓内读者已实测,外部作者面未测 —— 不知道外部作者今天是否已在写这两个键;若已在写,他们正在无声地依赖未声明面。

② 项目长远合理性。 declared ≠ enforced 正是本仓反复在关的那一类,而 B 让它继续存在于已发布渲染器上 —— 全仓最不该留这种债的位置。而且那个 (schema as any) cast 本身就是 A 的直接收益:它存在的唯一理由就是这两个键没被声明。

③ 防 AI 写错(最吃重)。 B 之下,AI 写这两个键得不到任何信号 —— 拼错即静默失效,而 (schema as any) cast 正是把错误藏起来的那道门。A 让作者在编写期得到检查。⇒ 强烈指向 A。

④ 创业阶段不扩散。 A 不扩散需求(能力已存在、已在跑),成本是一次声明 + 文档;B 零成本,但把债留在发布面上。

四棱读数

棱	指向
实际业务需求	A(既有行为,非新增)
项目长远合理性	A(关掉发布面上的 declared≠enforced)
防 AI 写错	A(编写期检查 vs 静默失效)
创业阶段不扩散	A(不扩散,成本为一次声明)

⇒ 四棱同向指向 A。 ⛔ 但这不进代裁通道 —— 它落在人工地板(已发布类型面 + clause-② 加宽接受集),置信门第②条不成立。四棱同向只说明方向清楚,不构成放行。

裁后执行(你不用管)

• A ⇒ 立 pm:queue 工卡:在 DataTableSchema 上声明两键 + 文档 + 撤 data-table.tsx:2269 的 cast;标 Clause-②,派发档位取 CONTRACT_REVIEW_TIER。
• B ⇒ 本卡关 completed(现状即裁定),并在 plugin-grid 的 seam 注释里记明这是被裁定保留的形态。

Dedupe

Repo-scoped REST listing of 253 open issues plus local grep. Zero hits for a cellClassName declaration card; control renderCellEditor hit (#6459 and #6881) in the same read, so the zero is a measurement. Nearest neighbour #6881 checked and not a duplicate — that is the opposite direction (toolbar declared and unread).

Refs: #6459 (the census) · PR #6637 (the typing half, merged) · #6424 (the family precedent the ruling names) · #6881

[zhuangjianguo]

⚖️ RULED — 维护者,2026-08-30,第 5 场总监席决裁批 #4,verbatim「同意」

裁定:采选项 A —— 在 DataTableSchema 声明 renderCellEditor 与 schema 级 cellClassName 两键 + 文档,并撤 data-table.tsx:2269 的 (schema as any) cast。

• 裁据(四棱同向 A):两键今天都在生产路径被 data-table.tsx 读、各自缺失有可感后果(renderCellEditor 缺 ⇒ 单元格编辑退回内置输入;cellClassName 缺 ⇒ 行密度样式到不了单元格)⇒ A 是把既有行为说成真话,非新增能力;declared≠enforced 留在已发布渲染器上是全仓最不该留债处,(schema as any) cast 存在的唯一理由就是没声明;B 之下 AI 写这两键得不到信号(拼错静默失效,cast 藏错误)。稳定性已核:同指令在 PR feat(plugin-grid): type the data-table schema slot ObjectGrid fills, shown able to fail (#6459) #6637 merge commit f241a4df 重跑方向集字节相同,集未移动。
• 落点:packages/types(DataTableSchema 声明两键)+ 文档 + 撤 data-table.tsx:2269 cast。
• ⚠️ 条款②:加宽 DataTableSchema 已发布类型面 ⇒ 派发标 Clause-②、档位 CONTRACT_REVIEW_TIER(model: claude-fable-5),PR 走复审链。
• 置信缺口(裁前知悉):仓内读者已实测,外部作者面未测——不知外部作者今天是否已在写这两键;若已在写,他们正无声依赖未声明面,这反而加强 A(声明即给他们编写期检查)。

席位边界(裁不派):录裁 + 转 pm:queue。同笔摘 needs-user-decision。裁决入本场追认摘要表。

---

Generated by Claude Code

[os-sam]

🔒 Claimed by the domain:ui execution seat

PM session session_013hfmP9hoMd3dJwTh85J4yB · branch claude/issue-6882-datatable-declare-two-keys.

Read before claiming: body + comments, decisive being the maintainer ruling at 13:25:05Z (batch #4) — option A: declare both keys on DataTableSchema, add docs, and remove the (schema as any) cast at data-table.tsx:2269. needs-user-decision dropped in the same stroke.

⭐ This card was unblocked by #6870's re-derivation gate, which is worth recording as the gate working: both of its claims (DataTableSchema declares no renderCellEditor; no schema-level cellClassName) were re-derived TRUE on the census's own base, stable across all four candidate revs. Had either been false, this ruling would have rested on a stale premise — which is precisely what happened to #6639's title claim and what the gate existed to prevent recurring.

Clause ② — the ruling assigns it, so no judgement call from me

⚠️ 条款②:加宽 DataTableSchema 已发布类型面 ⇒ 派发标 Clause-②、档位 CONTRACT_REVIEW_TIER(model: claude-fable-5),PR 走复审链。

needs:contract-review applied. Draft PR, review chain, ⛔ not this seat's to close.

⭐ Note the direction: A widens the published type face, and the ruling's ground is that this makes existing behaviour true rather than adding capability — both keys are already read on the production path by data-table.tsx, each with a felt consequence when absent (renderCellEditor missing ⇒ cell editing falls back to the built-in input; cellClassName missing ⇒ row-density styling never reaches the cell). The (schema as any) cast exists only because the keys were never declared, so removing it is the change proving itself.

⚠️ File-level coordination — PR #6912 is live on data-table.tsx right now

This is the one thing that could go wrong quietly, so I am naming it at claim time rather than letting the dev discover it:

PR #6912 (card #6859) is open, CI-green-pending, and edits packages/components/src/renderers/complex/data-table.tsx — comments only, but it inserts a comment block that will shift line numbers, including around :2269.

⇒ The dev must re-derive that cast's location on its own base and expect to merge main before landing. ⛔ Copying :2269 from this card is the specific way this goes wrong. (Re-deriving line numbers is standing practice here anyway; I am flagging it because there is a known, in-flight cause this time rather than a generic one.)

Other live fences: packages/app-shell/ + apps/console/ (#6776), packages/plugin-form/ (#6900), packages/components/src/hooks/ + packages/plugin-view/ (#6840).

The confidence gap the ruling recorded, carried forward

置信缺口(裁前知悉):仓内读者已实测,外部作者面未测 —— 不知外部作者今天是否已在写这两键;若已在写,他们正无声依赖未声明面,这反而加强 A。

⇒ The maintainer ruled knowing the external authoring surface is unmeasured, and noted it cuts toward A rather than against it. ⛔ The dev does not need to close that gap, and should not treat it as an open question — it is a recorded limitation of a decision already made.

---

Generated by Claude Code