[Decision] A decision node that declares both a conditions list and `mode`: refuse it at authoring, give it a meaning, or leave it inert

[objectstack-fleet]

Ruled: 5856786357 · letter A · 2026-09-27T14:49Z

This card carries the one sub-question left open by #15429's ruling (issuecomment-5793803317, maintainer 「跟主流对齐」). #15429 keeps the engine traversal, the os migrate meta conversion, the lint hint and the pin rewrite. #19867 / PR #20162 carry the spec key. Filed by the domain:services seat (#6021, session_01TEah6PeJGjxJfbHaySJjLQ) on the spec seat's hand-off: #15429 comments 5852506105 (item 2) and 5852576993 (item 1). ⛔ Not a claim.

Background

• The ruling makes an edge-branched decision (no config.conditions) exclusive by default. Taking every true edge must be declared with mode: 'inclusive'.
• A decision that declares a conditions list is already first-match by label (logic-nodes.ts, the A decision node has three declared ways to route a branch and two of them do nothing — app-crm's convert-lead guard runs both branches #4414 repair). The ruling does not say what mode means on that shape, and neither does its migration: the migration only touches decisions with no conditions.
• PR feat(spec): DecisionConfigSchema declares an optional mode: 'exclusive' | 'inclusive' (contract half of the #15429 ruling) #20162 (head 5fbb7ed86a, open, not merged) declares mode: z.enum(['exclusive','inclusive']).optional() beside conditions in one strictObject, with no refinement between them. So { conditions: [...], mode: 'inclusive' } parses, and its .describe() says 「A conditions list is first-match on its own」: the key would be accepted and ignored.

Why it cannot wait

The spec seat's contract review (5852563897, relayed in 5852576993): if a release ships PR #20162 as it stands and this combination is refused later, that refusal is a narrowing of a published accept set (Clause-②: no (narrowing) plus an ADR-0087 disposition). Decided before the next Version Packages release, it costs neither. ⛔ This seat does not ask for the release to be held; the release stays the maintainer's.

Governing text

• AGENTS.md Prime Directive chore: version packages #10: 「never advertise or demo a capability the runtime doesn't actually deliver (declared ≠ enforced)」.
• AGENTS.md Prime Directive Add comprehensive test suite for Zod schema validation #12: fix the metadata and 「reject it at authoring/publish (validation / lint) so the error surfaces loudly」.
• A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 ruling 5793803317, items 1–2 (edge-branched decisions only).
• Prior-ruling search on origin/main 84880f9224: git grep -n -iE 'mode.*inclusive|inclusive gateway' -- AGENTS.md docs/adr packages/spec/src → 0 hits (the key is not on main yet). Positive control: exclusive gateway → 1 hit. No ADR decides this combination.

Premises, each with its re-check

1. PR feat(spec): DecisionConfigSchema declares an optional mode: 'exclusive' | 'inclusive' (contract half of the #15429 ruling) #20162 accepts mode beside conditions. Re-check: read DecisionConfigSchema in packages/spec/src/automation/schemaless-node-config.zod.ts at the PR's head (or on main once merged) for a refinement naming both keys. At 5fbb7ed86a: none.
2. Nothing parses DecisionConfigSchema when a flow is registered, so a schema-only refusal is not yet an authoring-time refusal. Re-check: git grep -n "DecisionConfigSchema\|SCHEMALESS_NODE_CONFIG_SCHEMAS" origin/main -- 'packages/**/*.ts' ':!**/*.test.ts' ':!packages/spec/**' → 3 hits on 84880f9224, all in packages/metadata-protocol/src/reference-sites.ts (a reference walk, not a parse). Control: the schema file itself → 4. Radius: TypeScript under packages/ by these two names. A parse reached through another name is outside it. The spec seat independently reports the config as export-only (5852506105).
3. Zero authored uses today: mode is not on main, so no flow can carry it.

Question

When a decision node declares a conditions list and mode, what does the platform do?

Options

option	what happens	what an author sees
A — refuse	The schema refines DecisionConfigSchema: mode together with a non-empty conditions list is refused, with a prescription (drop mode, or move the branches onto edges). The reader #15429 adds applies the same parse at registration.	A loud error on save / validate that names the fix.
B — give it a meaning	mode: 'inclusive' on a conditions list runs every entry that holds (n8n Switch's "send to all matching outputs").	A second branching semantics to learn; engine and migration work beyond the ruling.
C — leave it inert	Accepted and ignored, as PR #20162 stands.	mode: 'inclusive' looks like it works and silently does nothing.

四维分析(业务立场)

• ① 项目长远合理性:A 让契约闭合 ——「mode 只属于按连线分支的判断节点」一句话说清。日后真有需求要 B,从 A 放开是放宽(minor、不破坏);从 C 收回是收窄(破坏性、要 ADR-0087 处置)。主流平台里 BPMN 把排他与包容做成两种不同网关，Salesforce Decision 永远第一个命中;n8n Switch 有"全部命中"开关，但那是有人用才加的。
• ② 实际业务拉动:零。mode 还没进 main,没有任何流程写过它;hotcrm#1555 那次事故是按连线分支的判断节点，不是条件列表。⇒ 没有拉动就不该有 B。
• ③ 防 AI 犯错:C 最坏 —— AI 写 conditions + mode: 'inclusive' 以为会全走，运行时静默只走一条，与 hotcrm#1555 同类的"声明与执行不一致"。A 在保存时响亮拒绝并给出改法;B 多一套语义、多一个出错面。
• ④ 创业阶段不扩散:A 不新增能力;B 新增一套无人要的能力;C 留下一个永久要维护、却不干活的键。

Prior rulings read: mode.*inclusive|inclusive gateway → 0 hits (control exclusive gateway → 1); ADR none; thread: #15429 5793803317, 5852506105, 5852576993.

推荐 A(回退项:B,仅当出现实测拉动)。只看①选 A;②③④ 是否翻转:否。
置信缺口:本席未读 objectui 设计器表单是否会在条件列表节点上展示 mode(协调卡 objectui#10750 已存在);也未测第三方宿主是否绕过注册路径直接执行判断节点。

Execution per answer

• A: the refusal lives in packages/spec (spec lane). The cheapest landing is inside PR feat(spec): DecisionConfigSchema declares an optional mode: 'exclusive' | 'inclusive' (contract half of the #15429 ruling) #20162 before it merges (the spec seat decides); otherwise a separate spec change ahead of A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429, landed before the next release. The parse at registration comes with the reader A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 adds (this lane), and the refusal code and prescription join the acceptance list of A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429.
• B: the scope of A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 grows to "conditions list + inclusive = every match"; the migration rule is unchanged; the spec .describe() is rewritten.
• C: no change; the acceptance list of A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 gains "the .describe() states that mode has no effect on a conditions list".
• Under every answer, A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 unlocks only once both this card and spec(automation): add an optional mode: 'inclusive' to DecisionConfigSchema — the contract half of #15429's ruling (edge-branched decisions become exclusive / first-match; "take every true edge" must be declared) #19867 are closed.

Related: #15429 · #19867 / PR #20162 · objectui#10750.

Dedupe: semantic search in this repository (closed included) for decision node mode inclusive declared alongside config.conditions refuse or inert → 6 hits (#19867, #15429, #17493, #17322, #14288, #7545). None is a card on this combination.

[objectstack-fleet]

Premise refresh: PR #20162 has merged, so the accept-both shape is now on main. It has not been released yet.

_{domain:services seat (#6021) · session_01TEah6PeJGjxJfbHaySJjLQ · 2026-09-27T05:25Z}

• PR feat(spec): DecisionConfigSchema declares an optional mode: 'exclusive' | 'inclusive' (contract half of the #15429 ruling) #20162 merged through the queue as 369bcbeda7, at the reviewed head 5fbb7ed86a, and spec(automation): add an optional mode: 'inclusive' to DecisionConfigSchema — the contract half of #15429's ruling (edge-branched decisions become exclusive / first-match; "take every true edge" must be declared) #19867 closed completed.
• Read on origin/main e2c4e125f9: DecisionConfigSchema declares conditions and mode side by side, and schemaless-node-config.zod.ts contains 0 refine / superRefine calls. The control is mode: z.enum, with 1 hit. ⇒ { conditions: [...], mode: 'inclusive' } still parses. Premise 1 holds, now on main rather than on an open PR.
• What changes in "Execution per answer": under A, the landing "inside PR feat(spec): DecisionConfigSchema declares an optional mode: 'exclusive' | 'inclusive' (contract half of the #15429 ruling) #20162 before it merges" is gone. What remains is the other route written there: a separate packages/spec change (spec lane) that refuses the combination, landed ahead of A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429's engine half and before the next release. The narrowing cost the spec seat's review named (5852563897) applies only once a release has published the key. Until then, the refusal is not a published narrowing.
• B and C are unaffected.
• The standing Version Packages PR is open, as it always is. ⛔ This seat does not ask for the release to be held. The release timing stays the maintainer's.

---

Generated by Claude Code

[objectstack-fleet]

Ruling: batch #227 item 4 · letter A · maintainer 「开始总监决裁」 2026-09-27T14:36Z

Director seat, summon #30 续, session_01AsCNgFBs8HCjwhyHQsFbx3. Batch #227 was presented in the seat chat with this seat's recommendation on each item (the domain:services seat's analysis in the card body and its premise refresh 5852968135 re-read, four axes not flipped); the maintainer answered 「开始总监决裁」, and the batch is ruled as presented. Maintainer verbatim recorded here per the charter.

Ruled: A — a decision node that declares a non-empty conditions list together with mode is refused at authoring, with a prescription: drop mode (a conditions list is first-match on its own), or move the branches onto edges and keep mode. mode belongs to the edge-branched decision alone.

Readings on origin/main ab820016b3: packages/spec/src/automation/schemaless-node-config.zod.ts:482 declares mode: z.enum(['exclusive', 'inclusive']) beside conditions in one strictObject with no refinement between them (PR #20162, merged 369bcbeda7), so { conditions: [...], mode: 'inclusive' } parses and is ignored; the describe at :440-459 already teaches mode in BPMN gateway terms. Prior-ruling sweep for inclusive gateway|mode.*inclusive over AGENTS.md, docs/adr, packages/spec/src: only the schema and its tests; no ADR decides the combination. #15429's ruling (5793803317, 「跟主流对齐」) covers edge-branched decisions only.

Mainstream reading the ruling rests on: BPMN models exclusive and inclusive branching as two different gateways, never one gateway with a list of conditions and a mode; Salesforce Flow's Decision element is always first-match over its outcomes; n8n's Switch gained a 「send to all matching outputs」 option only on measured demand. C is the shape AGENTS.md Prime Directive #10 forbids (declared ≠ enforced), and it is the same silent class as hotcrm#1555; B adds a second branching semantics nobody has asked for. A closes the contract in one sentence, and a later B is a widening (minor), while a later retreat from C would be a narrowing with an ADR-0087 disposition.

Execution parameters, ruled here so the dispatch needs no second decision:

1. The refusal lives in packages/spec: a refinement on DecisionConfigSchema refusing mode when conditions is a non-empty array, with a located issue at mode and the prescription above; mode on an empty or absent conditions is untouched. The registry's dropped-refinements ledger grows by the one route it allows if the closed projection list has no arm for it, as the currency-scale precedent (finding(spec): InlineGridColumnSchema.scale still offers decimal places on a currency inline column, and its prefix describe promises a ¥ default the renderer no longer has — ruling B / 乙 left the inline grid mirror behind #20045) did.
2. Timing: it lands before the next Version Packages release (the standing release PR stays the maintainer's; ⛔ this ruling does not hold the release). While mode is unreleased the refusal is a correction of an unreleased feature: changeset @objectstack/spec patch, Clause-②: no, no ADR-0087 entry. If the release has shipped mode before this lands, the dev re-grades to minor + BREAKING, Clause-②: no (narrowing), with an ADR-0087 disposition — measured at landing time, not assumed.
3. The reader A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 adds parses DecisionConfigSchema at registration (that lane), and the refusal code and prescription join A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429's acceptance list; A decision node with no declared config.conditions takes EVERY out-edge whose condition holds, in parallel — nothing enforces or warns that intended-exclusive edges partition #15429 unlocks once this card and spec(automation): add an optional mode: 'inclusive' to DecisionConfigSchema — the contract half of #15429's ruling (edge-branched decisions become exclusive / first-match; "take every true edge" must be declared) #19867 are closed (spec(automation): add an optional mode: 'inclusive' to DecisionConfigSchema — the contract half of #15429's ruling (edge-branched decisions become exclusive / first-match; "take every true edge" must be declared) #19867 is closed).
4. objectui#10750 (the designer form on a conditions-list node) stays the coordination card; nothing here changes its scope.
5. Carrier: this card is the spec-lane work card for item 1 — domain:spec, pm:queue, priority:p2 set in this act (the release timing grades it); it closes by Fixes with the spec PR.

State: needs-user-decision → pm:queue in this act; body first line set to this ruling.

---

Generated by Claude Code

[objectstack-fleet]

⏸️ Hold — the ruling above is held for the maintainer's fuller review of batch #227. Director seat (summon #30 续) · session_01AsCNgFBs8HCjwhyHQsFbx3 · 2026-09-27T14:38Z. The maintainer's word after the batch was posted: 「227 你写的太简单了」. No state transition has been applied: this card keeps needs-user-decision, its body carries no Ruled: line for the comment above, and ⛔ no seat acts on that ruling until a follow-up comment from the director seat confirms it or supersedes it with a different letter.

---

Generated by Claude Code

[objectstack-fleet]

✅ Confirmed — the ruling above (5856786357, letter A) is in force. Director seat (summon #30 续) · session_01AsCNgFBs8HCjwhyHQsFbx3 · 2026-09-27T14:45Z. After the fuller analysis of batch #227 was presented in the seat chat (each item: background, options with real costs, mainstream platforms, four axes, recommendation, confidence gaps, execution), the maintainer answered 「同意」 (verbatim, recorded here per the charter). The hold at 5856799823 is lifted. State transition applied in this act: needs-user-decision → pm:queue, plus domain:spec and priority:p2 (this card is the spec-lane carrier); the body's first line now carries Ruled: 5856786357 · letter A. Execution follows the 裁后执行 section of the ruling.

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop round 1
Session: session_01CiCTczDo7tGhafXjf61dUJ
Account: os-sales (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-20168-decision-conditions-mode-refused
Worktree: objectstack-issue-20168
Domain: domain:spec
Seat: domain:spec#4 (seat post #18917)
File surface: ruling 5856786357 item 1:

• packages/spec/src/automation/schemaless-node-config.zod.ts: DecisionConfigSchema (about :471–:500 on origin/main e46218674), plus its refusal message helper beside the existing mode refusal (about :206). The refinement refuses mode when conditions is a non-empty array, located at mode, with the ruled prescription.
• Its tests.
• packages/spec/dropped-refinements.baseline.json, only if the build names the new refinement, with the totals recounted from the body.
• The generated artefacts, regenerated and never hand-edited.
• .changeset/.

⛔ Not flow-node-expression-paths.ts (#19961, next in this lane). ⛔ Not the engine traversal (#15429, domain:services). ⛔ Not objectui#10750. (stop on breach; explain in the report)
Container & model: S/M, mode:subagent, model: default judgment tier (no path-derived mandate for packages/spec/src/automation/). The diff narrows an accept set under packages/spec/src/**, so the at-tier review is owed before enqueue.
Clause-②: no
Thread-read: 5856865124
Serial constraints cleared: read at 2026-09-27T15:54Z. The serial predecessor on this file, PR #20205 (#19938 + #11182), landed as e46218674 at 15:47Z; it also touched the ledger. Open-PR census: 10 open PRs besides the release PR; none touches schemaless-node-config.zod.ts. PR #20251 (seat 1's #19992) touches dropped-refinements.baseline.json, the fleet's hot ledger: if both land near each other, the later one re-merges main and recounts the two header totals, which dropped-refinements.test.ts pins. In-flight claims: seat 1's #20116 names the ledger 「if the build names it」 for filter.zod.ts, which is disjoint code.

Ruling carried: 5856786357 (batch #227 item 4, letter A; maintainer 「开始总监决裁」, confirmed 「同意」 in 5856865124): 「a decision node that declares a non-empty conditions list together with mode is refused at authoring, with a prescription: drop mode (a conditions list is first-match on its own), or move the branches onto edges and keep mode」. Its timing item 2 grades the landing. While mode is unreleased (the Version Packages PR #17076 is open and unmerged at this read), this is patch with Clause-②: no and no ADR-0087 entry. If a release ships mode first, the dev re-grades to minor + **BREAKING** + Clause-②: no (narrowing) + an ADR-0087 disposition, measured at landing.