[Decision] `DESIGN.md` §03 requires a termination reason but declares no field for it

[hotlong]

维护者速读

事情：设计方案第 7 章的状态转换守卫写着「生效 → 已终止：终止日期与原因必填」，但第 3 章的字段清单里没有「终止原因」这个字段。卡 02 实现状态机时只能盖终止日期，原因无处可存。

为什么现在问你：修法本身很小（加一个字段），但它要改 DESIGN.md §03，而仓库规矩是代码 PR 不得改 §01–§04，必须先有你的裁定。

选项：

• A（推荐）：加一个「终止原因」文本字段，只在终止时必填，其余时候不出现在表单上；在卡 09 落地，同时补 §03 字段清单。
• B：把守卫里的「原因必填」去掉，终止原因写在合同摘要里。

代价：A 多一个字段，多一行设计文档修订；B 不加字段，但终止合同将没有结构化的原因记录，审计时只能翻自由文本。

你要做的：回一个字母，A 或 B。

---

Context

DESIGN.md §03 state-machine table:

从	到	守卫
active	terminated	closed_at 与终止原因必填

The §03 field list for clm_contract declares closed_at but no reason field. Card 02 (#2, PR #5) implemented the guard as far as it could — it stamps closed_at and refuses without it — and reported the gap rather than inventing a field outside the pinned list.

Governing text: DESIGN.md §03 (the guard row above, and the clm_contract field list); AGENTS.md → Delivery process → "Files a code PR never touches: … DESIGN.md §01–§04 without a needs-user-decision first".

Options

A — declare termination_reason (textarea, requiredWhen the status is terminated), amend the §03 field list in the same change, implement in card 09 alongside F13/F14. The field never appears on the intake form; it is asked exactly once, at termination.

B — drop "原因必填" from the guard wording. The reason, when someone bothers to record it, lives in summary.

四棱分析

实际业务需求 — A. 提前终止的合同是法务与审计最常回头查的一类记录，而「为什么终止」正是他们问的第一个问题：是对方违约、是我方不再需要、还是双方协商。设计方案自己的守卫已经写了「原因必填」，说明写方案时就认定这是业务需要，不是事后加戏。B 把这个答案散进自由文本，等于放弃了按原因统计终止合同的能力。

项目长远合理性 — A。当前设计文档自相矛盾：守卫引用了一个字段清单里不存在的字段。两个选项都能消除矛盾，但方向相反：A 让实现追上已声明的意图，B 削弱意图去迁就实现。契约优先的规矩是以协议为基准、改代码对齐，A 是那个方向。

防 AI 写错 — A，而且这一棱最不平衡。requiredWhen 是运行期强制的结构性约束：任何写入者，无论是人、流程还是 AI，都不可能在不给原因的情况下终止一份合同。B 依赖的是约定（「记得写进摘要」），而约定恰恰是错误能静默累积的地方 —— 这正是「声明即强制，绝不让 AI 声明一个运行时不兑现的能力」要防的形状。今天守卫说必填而系统不存原因，就已经是一个不兑现的声明。

创业阶段不扩散 — 大致中性，略偏 A。A 只加一个字段，且这个字段是已确认的 V1.0 基准自己要求的，不是新增能力面，因此不构成需求扩散。B 表面上更紧，但它删掉的是一条已声明的能力，属于反方向的基准变更，同样需要你拍板。

结论：四棱同向 A，第三棱（防错）差距最大。推荐 A。

Impact if deferred

None blocking. Card 02 shipped with closed_at only; card 09 (F13/F14, expiry and archive) is where the terminate path gets its flows and is the natural place to land the field. Answering before card 09 is dispatched avoids a second pass over the same hook.

[zhuangjianguo]

裁定：A —— 加 termination_reason 字段

维护者裁决，2026-09-09，常设决裁批第 1 批，逐字：

第 1 批 平台的问题去平台修，业务的问题按照你的意见。

本卡是业务问题（设计方案第 3 章内部自相矛盾：守卫引用了一个字段清单里不存在的字段，与平台能力无关）⇒ 按席位推荐 A。

Governing text（录裁前重跑）：DESIGN.md §03 状态机表 active → terminated 行「closed_at 与终止原因必填」+ clm_contract 字段清单；AGENTS.md → Delivery process →「Files a code PR never touches: … DESIGN.md §01–§04 without a needs-user-decision first」。该前置条件现已满足。

鲜度门：正文最后编辑 2026-09-07T12:34:55Z，其后本卡零评论。无更新前提，裁定录在当前正文之上。

裁后执行

1. §03 编辑（受管面，随折叠的设计文档 PR，链首本卡）：clm_contract 字段清单增加 termination_reason，textarea，requiredWhen 状态为 terminated；不出现在发起表单上，只在终止那一刻问一次。
2. 代码落地：随卡 09（F13/F14 到期与归档）—— 那是终止路径拿到流程的地方，也是本卡「Impact if deferred」段自己指定的落点。
3. 设计文档 PR 触受管面 ⇒ 按 AGENTS.md 由席位 ACCEPT 后留给维护者合并，附 ## 维护者速读；代码 PR 走自主路径。

状态转换：needs-user-decision → pm:queue（同一笔标签写入）。

⚠️ 置信缺口照录：我呈报时说过，看不见你们要的是自由文本还是封闭选项清单。裁定是 A（加字段），字段形态默认自由文本 textarea，因为那是卡片正文写明的形状。若你要的是可统计的封闭清单，现在说一句，落地形态改一行；等 §03 落了地再改就是一次迁移。

---

Generated by Claude Code

[zhuangjianguo]

Claim: PM loop round 2 (seat repo:hotclm) — chain head of a folded family dispatch (#6 · #10 · #14, with #31 as a declared rider). Dispatched by the PM seat for its dev; the dev inherits this claim and posts no second one.

Session: session_01R3n3GGzobdegM4HUzah1iR
Branch: claude/issue-6-design-rulings (shared; named for the chain head)
Worktree: hotclm-issue-6 (dedicated, off origin/main @ 5d5a254)
Domain: repo:hotclm (single-lane repo — no domain:* axis)
File surface: DESIGN.md only. ⛔ Nothing under src/, scripts/, or anywhere else (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: default judgement tier — small, but every edit lands on a governed surface and each must match a ruling verbatim, so ⛔ not the floor tier.
Clause-②: no — recording adjudications already made in the design document neither relaxes an acceptance set nor widens a published contract surface.
Thread-read: 5604405032 (this seat's ruling comment on #6, 2026-09-09T15:0xZ)
Serial constraints cleared: #39 (card 09, branch claude/issue-39-post-signature-jobs) is dispatched in the same batch and is the only sibling in flight. Disjoint by construction: #39 owns src/, this owns DESIGN.md, and each dispatch order forbids crossing. Zero other pm:dispatched cards, zero open PRs at 2026-09-09T15:0xZ.

Folding gates — all five, answered

• ① same defect shape, same fix. Every member is "a ruling landed today and the design document still says the old thing". The fix is identical in each case: one editorial amendment recording an adjudication. ⛔ Not merely "same subsystem" or "same keyword".
• ② same package / region. DESIGN.md, one file, one worktree, one PR, one queue slot.
• ③ every member already adjudicated. [Decision] DESIGN.md §03 requires a termination reason but declares no field for it #6 → A, [Decision] Two states in DESIGN.md §03 are dead ends: a started obligation cannot go overdue, a partly paid instalment cannot be completed #10 → 1A+2A, Decision: where does the intake-captured payment arrangement F9 reads actually live? #14 → A, [Decision] §09 要「各段时长」和「超 SLA」，但语义层算不出时间差——两问共用一个补法 #31 → 2B — all ruled 2026-09-09 by the maintainer. ⛔ No card from the decision box is in this fold.
• ④ every member independently checkable. Each has a named section and a named ruling: §03 field list ([Decision] DESIGN.md §03 requires a termination reason but declares no field for it #6) · §03 two state-table rows ([Decision] Two states in DESIGN.md §03 are dead ends: a started obligation cannot go overdue, a partly paid instalment cannot be completed #10) · §06 F9 (Decision: where does the intake-captured payment arrangement F9 reads actually live? #14) · §09 ([Decision] §09 要「各段时长」和「超 SLA」，但语义层算不出时间差——两问共用一个补法 #31's 2B).
• ⑤ exclusion list — what looks like family and is not:
  • [Decision] DESIGN.md §04 makes clm_requester every employee's default, but a set that grants clm_requester.access cannot bind to everyone #11 — ⛔ excluded, and this is the one that matters. Its §04 sentence would fit this PR perfectly, but [Decision] DESIGN.md §04 makes clm_requester every employee's default, but a set that grants clm_requester.access cannot bind to everyone #11 still carries needs-user-decision because its second question is unruled, and the red line is absolute: never dispatch a card carrying that label. Its §04 sentence lands after question 2 is answered.
  • Card 09 — F9 activation · F3/F4/F10–F14 scheduled jobs · F16 executed upload (M3) #39 / card 09 — the code half of the very same rulings. Excluded on file surface, not on subject.
  • Decision: DESIGN.md §10 asks for 300 seeded contract versions, but §03 makes file required and a seed cannot mint one #19 · Decision: §10 wants a multi-entity group, but clm_contract.our_entity ships exactly one option and a seed cannot add one #20 · [Decision] pnpm demo 开箱即用时没有任何账号能打开「我的合同」——要不要让脚本建账号，与 #11 第一问耦合 #28 · [Decision] 17.4 的新规则指出 21 个字段声明了却没人读——展品该给它们消费者，还是删掉声明 #38 — still in the decision box (batch 2).
  • [Decision] §09 要「各段时长」和「超 SLA」，但语义层算不出时间差——两问共用一个补法 #31 itself — carries pm:blocked (Blocked-by: objectstack-ai/objectstack#16737) and is ⛔ not being dispatched. Only the §09 wording correction its own 2B ruling ordered rides here, as Refs #31, ⛔ never Fixes — the card stays blocked until the platform ships.

Landing path — governed, so it does NOT self-merge

DESIGN.md §01–§04 is a governed surface under AGENTS.md. This PR is ACCEPTed by the seat and then left open with a ## 维护者速读 for the maintainer to merge. ⛔ The seat's merge delegation does not cover it. The dev writes the 速读 draft in the PR body (Chinese, business framing, seat opinion left blank); the seat corrects it against its own reading of the diff and posts the final version.

Members carry Refs, ⛔ not Fixes: the substantive half of #6/#10/#14 is code, and that lands in #39. These cards close when #39 lands, not here.

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 6,
  "family": [6, 10, 14, 31],
  "status": "done",
  "branch": "claude/issue-6-design-rulings",
  "pr": "https://github.com/objectstack-ai/hotclm/pull/40",
  "premise_still_valid": true,
  "summary": "Four maintainer rulings from 2026-09-09 batch 1 recorded in DESIGN.md, one file, one commit (1127e52). #6 ruling A: §03 clm_contract field list gains termination_reason (textarea, requiredWhen status is terminated, absent from the intake form), so the active-to-terminated guard now names a field that exists. #10 rulings 1A+2A: clm_obligation.status gains in_progress to overdue; clm_payment_plan.status gains partial to paid and partial to overdue. #14 ruling A: §06 F9 loses the clause about generating clm_payment_plan from an intake-captured payment arrangement, with no replacement sentence and no §03 field added. #31 ruling 2B (rider): §09 restates contract_cycle_time as stage-coverage counts, replaces the by-type over-SLA promise with the fixed 30-day threshold whose value is stated in the widget title, replaces average turnaround with approval throughput via the platform compareTo primitive, and records that both original metrics wait on objectstack-ai/objectstack#16737. All four dispatch-quoted strings matched the file exactly before editing.",
  "tests": "Four gates run in the worktree on 1127e52, each exit code captured before any pipe via 'cmd > log 2>&1; EXIT=$?': pnpm validate=0, pnpm lint=0, pnpm typecheck=0, pnpm lint:i18n-gate=0. None moved, and structurally none could: grep over objectstack.config.ts, scripts/, tsconfig.json, package.json and .github/workflows/ finds DESIGN.md only inside code comments, and tsconfig include is objectstack.config.ts plus src only, so no gate parses the edited file. Pre-existing warnings in validate and lint (clm_review.comments, clm_review.internal_note, clm_signature.envelope_id/signers/notes carrier-only; gm_signoff empty-slate suggestion) all name src declarations this PR does not touch; lint reports 21 warnings and 5 suggestions, unchanged in kind. i18n gate: 0 missing keys across 2 locales, 1525 keys expected. No ablation applies: this diff adds no gate and no test. No browser verification performed or needed, stated explicitly because the reader would otherwise wonder: the diff edits a design document that no gate parses and no runtime loads, so there is no runnable surface and nothing to click.",
  "files_changed": ["DESIGN.md"],
  "gates": {"validate": 0, "lint": 0, "typecheck": 0, "lint:i18n-gate": 0},
  "deviations": [],
  "mcp_calls": "0 — every GitHub read and write went over repo-scoped REST, probed first (GET repos/objectstack-ai/hotclm returned 200, public repo)",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: §02 line 80 cross-references §09 as 'Insights | 周转分析 | 阶段时间戳 → dataset → 看板（§09）'. After this PR §09 declines per-stage durations, so §02 still advertises turnaround analysis and points at the section that now says it is not delivered. This is exactly the cross-reference the dispatch asked me to look for; §02 is outside the ruled surface (Zone 1) so it is untouched. Successor: the PM seat reviewing PR #40, and #31 which owns this defect class.",
    "noted, not filed: §09's 管理层 board still lists 审批瓶颈（各台阶平均停留）, a third per-stage duration of the same unbuildable shape as the two #31 ruled on, named by neither #31's body nor its ruling. Verified against delivered code: src/dashboards/executive.dashboard.ts ships four 'Routes:' load-count widgets instead, and its own comment records that sys_approval_request holds 0 rows on a stock demo, so there is no dwell to read. §09 is therefore now correct about two of three duration promises and still wrong about the third. Closing it is a wording call of the same kind as #31's 2B and belongs to the maintainer, not to a PR whose warrant is to record what was ruled. Successor: the PM seat reviewing PR #40.",
    "noted, not filed: §09's known-limits line cites gap '#10' but §12's 编号 column carries only #9, #39, #27, #1, #7 and #36 — no #10 row. Pre-existing, cosmetic, untouched. Successor: whoever next edits §12."
  ]
}

---

Generated by Claude Code

[claude]

os-dev-report

(rework round 1 — supersedes the report of 2026-09-09 earlier on this card)

{
  "issue": 6,
  "family": [6, 10, 14, 31],
  "status": "done",
  "rework_round": 1,
  "branch": "claude/issue-6-design-rulings",
  "pr": "https://github.com/objectstack-ai/hotclm/pull/40",
  "head": "9e53a01",
  "premise_still_valid": true,
  "summary": "Rework round 1 addressed, both edits landed on the same file, branch and PR — no new PR. (1) §09's 管理层 bullet no longer reads 审批瓶颈（各台阶平均停留）, which was itself a per-stage duration and contradicted the exclusion paragraph three lines below it; it now states what the board actually ships — four routing-load tiles, one per rung of the §04 ladder, reading the route_* flags F2 stamps, described as 流量 and explicitly not 停留. The tile is described, not deleted. The limits paragraph now also names the dwell and records its second, non-platform cause: sys_approval_request holds 0 rows on a stock demo and the F5 ladder never ran, so the rungs have no start and no end to subtract — a cause that outlives objectstack-ai/objectstack#16737, stated so no reader expects the platform fix to deliver a dwell. The closing prohibition now names 各段时长 and 超 SLA instead of saying 这两项, which became ambiguous once a third metric was discussed in the same paragraph. (2) §02's Insights row no longer points at §09 for 周转分析; it now says stage stamps deliver 各阶段到达数与审批吞吐 with durations waiting on the platform. Exactly one line of §02 moved, per the widened scope. I accept the coordinator's boundary call: a §09 that both excludes durations and promises one is not truthful, so making it consistent is ruling 2B applied rather than extended, and shipping the dangling §02 pointer for later filing would have been the push-it-downstream pattern these cards exist to stop. Round-1's four ruled edits are unchanged.",
  "tests": "Four gates re-run in the worktree on 9e53a01, each exit code captured before any pipe via 'cmd > log 2>&1; EXIT=$?': pnpm validate=0, pnpm lint=0, pnpm typecheck=0, pnpm lint:i18n-gate=0. Identical to round 1 on 1127e52 down to the counts — lint still reports 21 warnings and 5 suggestions, i18n gate still 0 missing keys across 2 locales over 1525 expected. Structurally none of the four could move: DESIGN.md appears in objectstack.config.ts, scripts/ and .github/workflows/ only inside code comments, and tsconfig include is objectstack.config.ts plus src/**/* only, so no gate parses the edited file. Pre-existing validate/lint warnings all name src declarations this PR does not touch. CI on 9e53a01: 2 check runs, both Validate, both completed success — read once, not polled. Hunk map of the whole diff against origin/main confirms the authorized surface only: line 80 (§02, the single rework-authorized line), 106 (§03 field list), 140-141 (§03 state rows), 234 (§06 F9), 297/302/307-308 (§09); §01, §04, §05, §07, §08, §10, §11, §12, §13 untouched. Control-byte scan of DESIGN.md clean (grep -naP over the C0 range plus DEL, exit 1). No browser verification performed or needed: the diff edits a design document no gate parses and no runtime loads, so there is no runnable surface to drive. The two rework edits do not even touch code — src/dashboards/executive.dashboard.ts already shipped the four routing-load tiles; the edit makes the document match already-released behaviour.",
  "files_changed": ["DESIGN.md"],
  "commits": ["1127e52 — the four ruled edits", "9e53a01 — rework round 1, the two consistency edits"],
  "gates": {"validate": 0, "lint": 0, "typecheck": 0, "lint:i18n-gate": 0},
  "ci": "2/2 green on 9e53a01 (Validate x2, both success)",
  "pr_body_updated": "改了什么 and 风险与代价 rewritten to cover both new edits; 席位意见 verified still blank in the stored body; Refs lines unchanged, still no Fixes",
  "deviations": [],
  "mcp_calls": "0 — every GitHub read and write went over repo-scoped REST, probed first (GET repos/objectstack-ai/hotclm returned 200, public repo)",
  "open_questions": [],
  "out_of_scope_findings": [
    "knowingly left, per the coordinator's instruction: §09's known-limits line cites gap '#10' but §12's 编号 column carries only #9, #39, #27, #1, #7 and #36 — no #10 row. Pre-existing, cosmetic, unrelated to any of the four rulings, so it is left as found rather than fixed as a rider. Successor: whoever next edits §12.",
    "resolved in this round, no longer open: §02's dangling Insights pointer and §09's contradictory 管理层 bullet — both were round-1 findings and both were promoted into the diff by the rework."
  ]
}

---

Generated by Claude Code

[zhuangjianguo]

Review: ACCEPT — PR #40, head 9e53a01, CI 2/2 Validate completed/success. ⛔ Not merged and not mergeable by this seat: the diff is DESIGN.md, a governed surface, so it is flipped ready and left for the maintainer with a final ## 维护者速读 comment. Readings below are mine, taken against the tree and GitHub, ⛔ not read off the report.

Independent verification

Check	Method	Result
Path surface	git diff --name-only origin/main...FETCH_HEAD	DESIGN.md, and nothing else
Governed hit	grep of that list against the AGENTS.md governed set	1 hit ⇒ maintainer-merge path, not the seat's
Head binding	git rev-parse FETCH_HEAD vs the PR object's head.sha	both 9e53a01e7703…; the check runs are bound to that head, ⛔ not to the SHA an event carried
Hunks vs claim	git diff -U0 … | grep '^@@'	lines 80 · 106 · 140–141 · 234 · 297 · 302–303 · 308 — §02, §03, §06, §09 only. §01, §04, §05, §07, §08, §10–§13 untouched, as claimed
Diff read	line by line against the four ruling comments	each edit matches its ruling; no scope creep, and no replacement sentence keeping a deleted idea alive — #14's ruling A took a pure deletion, which is the part most likely to have been softened and was not

mergeable_state: clean, 1 file, +10/−8, 2 commits.

Round 1 → REWORK, and what came back was better than what I asked for

Round 1 recorded all four rulings correctly, but left §09 contradicting itself: the new paragraph declared per-stage durations out of scope while the 管理层 bullet three lines above still promised 「审批瓶颈（各台阶平均停留）」 — itself a per-stage duration. The report classified that as a maintainer wording call. I disagreed and said so: ruling 2B is 「改 §09 措辞」 to make §09 truthful, and a section that contradicts itself is not truthful, so fixing it applies the ruling rather than extending it.

The rework did that, and then added something I had not asked for and should have: it found a second, platform-independent reason the metric cannot exist. Approval-step dwell has no readable start and end because sys_approval_request holds zero rows on stock demo data — the F5 ladder has never run. So even after the platform gains date arithmetic, that number still cannot be computed. §09 now says so explicitly, which stops the next reader assuming a platform fix restores it. That is the difference between recording a ruling and understanding it.

The bullet was rewritten to what the board actually ships — four route-volume tiles reading F2's route_* flags, with 「是流量，不是停留」 stated at the site — rather than deleted. Correct: there is a real tile there.

⛔ I widened scope by one line, deliberately, and it is on the record

My Zone 1 forbade touching §02. I relaxed it for exactly one line in the rework: §02's Insights row pointed at §09 for 周转分析, and this PR is what made that pointer dangling. Shipping the contradiction and filing it for later is the "push the conflict downstream" pattern these very cards exist to stop. One line, named in the rework instruction, and verified in the diff as one line — §02 line 80 and nothing else in §02.

Findings accepted as left

§09 cites gap #10, and §12's 编号 column carries #9, #39, #27, #1, #7, #36 — no #10. Pre-existing, cosmetic, unrelated to any ruling. Correctly left as found rather than smuggled in as a rider.

Why no browser evidence, and why that is right

AGENTS.md scopes its browser rule to 「every card that changes a surface a human touches」. This diff edits a design document no gate parses (tsconfig.include is objectstack.config.ts + src/**/*) and no runtime loads. Stating that explicitly, rather than silently omitting it, is the correct handling.

Card status — ⛔ this PR closes nothing

Refs, never Fixes, and that is right: the substantive half of #6, #10 and #14 is code, and it lands in #39 (card 09, in flight now). #31 stays pm:blocked on objectstack-ai/objectstack#16737. All four cards stay open until #39 lands.

---

Generated by Claude Code