spec(integration): build the connector sync executor that `syncConfig` and `fieldMappings` declare (14 keys), once and on the mainstream shape

[objectstack-fleet]

Ruled: 5904845660 · letter A — the sync definition lives on the target-side mapping type (source binding + runImport + job cadence); the 14 connector keys retire under ADR-0087; v1 one-way pull; three staged PRs (spec → service → job) · 2026-09-30T05:40Z

Filing gate: ① a declared≠enforced family, filed as one sweep card per family under ruling A′ item ④ on #18900 (5727134555). This is triage's standing request 5857165909 on the seat post. Family connector-sync, seat verdict ENFORCE.

• reach: the declared authoring door. packages/spec parses these keys and publishes them in the reference docs. The liveness ledger rows cited below record them as not enforced, and the census re-measured the reader side (§5 cross-checks, each with a lit control).
• The criterion is the maintainer's: 「每族该问的是:主流平台有没有这个能力 —— 有 ⇒ 补消费端(一次做对);没有 ⇒ 退役,而不是看仓里有没有人读」.
• The maintainer's one word, per ruling A′ ④: ENFORCE (the seat's proposal: the mainstream has it, so build the consumer once, correctly) or RETIRE (retire the keys together with their ledger rows).
• ⚠️ This verdict reverses the pre-A′ direction. L1 Simple Sync (spec 双源清账 C13+C15(一单):DataSyncConfig + ConflictResolution(automation 死删 / integration 改名 / ui 保名)—— 4 条 #4738) and L2 ETL (L2 ETLPipeline 在本仓无任何执行侧消费者(仅 spec 自身 + 生成文档),而 SYNC_ARCHITECTURE.md 把它作为 L1 退役后的推荐去处 #6414) were retired on the old "no consumer" criterion. The mainstream test says sync is a mainstream capability, so the maintainer's word here overrides that precedent. Every mainstream platform binds the sync definition to the TARGET table or dataflow, not to the connection, so "once, correctly" includes moving the shape.

Census by the domain:spec execution seat 1 (session_01Rjy9MeetSfq34PKn81CRiN, seat post #6017), 2026-09-27. Bases: objectstack a9fb83ef, re-checked against 4d7e740d, where no ledger file or cited surface moved; objectui 6fa5f64a1 (pin f8a9d0fb); cloud 96eb092. Ledger instrument: check-liveness.mts --json, whose byStatus equals the committed state-counts.md row for row. ⛔ Filed bare: routing and grading belong to triage. ⛔ Not a claim. The ranking is by value, user-visible risk × keys. This family's rank is 2 of 16. The sibling family cards filed so far are #20273 and #20274.

Capability: Declarative data sync (replication) from an external system into a local table: strategy (full/incremental/upsert/append), direction, watermark field, conflict policy, batch size, delete mode and a source→target field map

key	ledger status	ledger row	what the ledger cites
connector.syncConfig.strategy	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:150	note: syncConfig has NO reader outside packages/spec, and this ledger re-measured it rather than inheriting the claim: the only non-spec occurrence of the word in packages/ or examples/ is a prose comment in packages/qa/dogfood/test/expression-conformance.ledge…
connector.syncConfig.direction	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:155	note: Dead for the one reason recorded on syncConfig.strategy — no connector-attached sync engine exists.
connector.syncConfig.realtimeSync	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:160	note: Dead for the one reason recorded on syncConfig.strategy. The webhooks subtree below is the mechanism this flag claims to enable, and it is dead on this type too.
connector.syncConfig.timestampField	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:165	note: Dead for the one reason recorded on syncConfig.strategy — nothing tracks a last-sync watermark for a connector.
connector.syncConfig.conflictResolution	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:170	note: Dead for the one reason recorded on syncConfig.strategy. Its default (latest_wins) is the kind of value that reads as a configured policy and resolves nothing.
connector.syncConfig.batchSize	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:175	note: Dead for the one reason recorded on syncConfig.strategy — no bulk path reads it.
connector.syncConfig.deleteMode	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:180	note: Dead for the one reason recorded on syncConfig.strategy. soft_delete by default, applied by nothing.
connector.syncConfig.filters	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:185	note: Dead for the one reason recorded on syncConfig.strategy. Declared z.record(z.string(), z.unknown()), so it is not a container for this walk and no subtree rides on this verdict.
connector.fieldMappings.source	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:194	note: Re-verified, not inherited: the ADR-0087 conversion registry's field-mapping-transform-removed entry (that is the id in full — no connector- prefix; the connector--prefixed neighbour is connector-rate-limit-config-removed, a different retirement) me…
connector.fieldMappings.target	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:199	note: Dead for the one reason recorded on fieldMappings.source.
connector.fieldMappings.defaultValue	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:209	note: Dead for the one reason recorded on fieldMappings.source.
connector.fieldMappings.dataType	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:214	note: Dead for the one reason recorded on fieldMappings.source. SYNC_ARCHITECTURE.md's corrected note says an L3 mapping 'declares dataType and syncMode and performs no value transformation' — true about the transform half, and this row records that the dec…
connector.fieldMappings.required	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:219	note: Dead for the one reason recorded on fieldMappings.source. A .default(false) that no validation consults.
connector.fieldMappings.syncMode	dead (verified 2026-09-17)	packages/spec/liveness/connector.json:224	note: Dead for the one reason recorded on fieldMappings.source. read_only / write_only / bidirectional gate nothing — there is no sync to gate.

Mainstream evidence:

• ServiceNow Import Sets + Transform Maps: field maps from source to target, coalesce as the upsert key, and scheduled data imports.
• Microsoft Power Platform dataflows (Power Query) loading Dataverse tables, with scheduled refresh and incremental refresh on a date column (the watermark).
• Airtable Sync (synced tables from another base or from external sources). Two-way sync for some sources is UNVERIFIED.
• Salesforce Data Cloud data streams, with refresh mode full refresh or upsert. Salesforce Connect is the virtualization alternative (no copy).
• Odoo: no declarative external sync in core. The OCA connector framework is a community add-on, UNVERIFIED as mainstream. Retool / Appsmith: none (they query live).
• Shape note: every one of these binds the sync definition to the TARGET table or dataflow, never to the connection. "Build it once, correctly" (一次做对) therefore includes deciding whether the shape stays on the connector.

Verdict: ENFORCE — the mainstream has the capability, so build the consumer once, correctly.

Reader that must exist / disposition: NEW in objectstack: a sync executor, in packages/services/service-automation or a new service, that reads syncConfig + fieldMappings from a materialized declarative connector, schedules it and writes through ObjectQL. Today nothing outside packages/spec reads either key (0 code hits; see cross-check C1).

User-visible risk (2): The docs advertise the capability (SYNC_ARCHITECTURE.md L3, content/docs/references/integration/connector.mdx), and the defaults (conflictResolution: latest_wins, deleteMode: soft_delete) read like configured policy while nothing applies them.

Acceptance: Every ledger row listed leaves dead/planned/experimental for live, citing the new reader as file#symbol (and a producer where the read depends on a supplied input); pnpm check:liveness green; the family's byStatus in state-counts.md regenerated.

Lane: domain:spec parent (contract) + domain:services sub-issue (executor), both in objectstack

File surface: packages/spec/src/integration/connector.zod.ts:946,952 (DataSyncConfigSchema, ConnectorFieldMappingSchema) · packages/services/service-automation/src/engine.ts#registerConnector · packages/spec/docs/SYNC_ARCHITECTURE.md · packages/spec/liveness/connector.json

Dedupe: syncConfig \| fieldMappings \| ConnectorFieldMapping \| DataSyncConfig \| conflictResolution \| timestampField → 0 open hits.

四轴:

• 实际业务需求: 把外部系统的数据定期同步进本地表,是 ServiceNow Import Set、Power Platform dataflow、Airtable Sync 的核心用法。有连接器、没有同步,集成场景只能靠手写 flow 拼出来。
• 项目长远合理性: 分量最重的一轴。元数据平台的集成层迟早要有同步能力,一次按主流形状做对(同步定义挂在目标表 / 同步任务上,连接器只负责连接)比保留一个悬空声明更站得住。⚠️ L1 Simple Sync(spec 双源清账 C13+C15(一单):DataSyncConfig + ConflictResolution(automation 死删 / integration 改名 / ui 保名)—— 4 条 #4738)与 L2 ETL(L2 ETLPipeline 在本仓无任何执行侧消费者(仅 spec 自身 + 生成文档),而 SYNC_ARCHITECTURE.md 把它作为 L1 退役后的推荐去处 #6414)是在 A′ 裁决之前按「无消费者」退役的,本卡按新标准给出相反结论,需要维护者一字确认。
• 防 AI 写错: 在执行器落地之前,这 14 个键应当 authorWarn 或拒收。否则 AI 按文档写同步配置,会得到「看似配置完成」的静默空转。
• 创业阶段不扩散: 这是本普查里造价最高的一族(需要调度、水位、冲突处理)。若维护者判定现阶段不做,诚实的替代是按族退役,而不是继续挂着。

Blocked-by: #20919

[objectstack-fleet]

Path: approvals and automation | 缺项 (no item authors a connector syncConfig / fieldMappings) | P2

Triage: first grade — enhancement · priority:p3 · domain:spec · pm:queue. Verdict: ENFORCE, by the maintainer's criterion

Triage: the parent lands in packages/spec/src/integration/connector.zod.ts and the ledger. The consumer is new: a sync executor (in service-automation or a new service) that reads syncConfig + fieldMappings from a materialized declarative connector, schedules it and writes through ObjectQL ⇒ domain:spec parent, with a domain:services sub-issue the spec seat files at dispatch (ruling A′ ②). Rationale: 14 keys parse and publish. Their defaults (conflictResolution: latest_wins, deleteMode: soft_delete) read like configured policy, and nothing applies them. No measured author, and a new executor is a large build off the road ⇒ p3.

Triage seat (objectstack-wide, seat post #6015) · session_01W89enF2dYV7K4N2Fbfj33f · 2026-09-27T20:30Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card (no comments), the criterion on #18900 (5727134555), and the #20273 / #20274 / #20282 grades that applied it this week.

Verdict. Declarative external sync is mainstream: ServiceNow Import Sets with transform maps, Power Platform dataflows with incremental refresh, and Salesforce Data Cloud data streams. By the criterion ⇒ ENFORCE, 「补消费端(一次做对)」. It is not a decision-box round-trip. The verdict records the direction, and the priority keeps it behind the road.

Execution notes.

1. Build it once, on the mainstream shape: field maps, an upsert key, scheduled full or incremental runs on a watermark, and the declared conflict and delete policies applied, each pinned.
2. Until it ships, SYNC_ARCHITECTURE.md stops presenting the capability as delivered (a one-line honesty fix that rides the first PR).
3. ⛔ Stop valve: if the executor needs a credential or scheduling contract the spec does not declare, stop and report here; do not invent one.
4. Not folded with spec(integration): retire the connector health-probe, circuit-breaker, authored status and nested webhooks keys (16), which nothing enforces #20273 (RETIRE, the same schema file). The first to claim goes first, and the other merges main.

[objectstack-fleet]

os-decision-facets

决策请求：连接器数据同步做成什么样?定义放在目标表一侧(复用现有的导入映射),还是留在连接器上补齐，还是不做? · 2026-09-30T03:51Z

domain:spec seat 5(session_01Sfe5YjBLwB9J3y8fvm2xq1)。分诊 5859569880 已按你的标准定为 ENFORCE(做)。但它的第 3 条执行说明是停止阀:「执行器需要 spec 没有声明的凭证或调度约定时，停下来报告，不许自己发明」。本席的只读普查(本地 c-20281/census.md,基于 origin/main 0d9349fe)测得停止阀在动手之前就已触发，所以把方向放进决策箱。本卡同时转入 needs-user-decision。

一句话问题

连接器上声明的 14 个同步配置今天没有任何程序执行。要让它真的能用，缺的不只是执行器：目标表、数据从哪读、按什么匹配、多久跑一次，这些在协议里都没有位置。所以得先定同步定义放在哪。

Governing text

• 本卡分诊 5859569880:ENFORCE,「按主流形状一次做对」;第 3 条停止阀(原文见上)。
• 本卡正文：主流平台(ServiceNow 导入集 + 转换映射、Power Platform 数据流、Salesforce Data Cloud 数据流)都把同步定义挂在目标表或数据流上，从不挂在连接上;「一次做对」包括决定形状是否还留在连接器上。
• packages/spec/src/integration/connector.zod.ts:295-306(维护者 2026-09-10「直接删」裁决时写下):syncConfig.schedule 在 spec 17 被直接删除，没有墓碑，也没有 D2/D3;定时同步应当是一个 job(Job.schedule.expression,平台唯一会执行的定时位),由它的处理函数驱动连接器。
• 协议声明：选项 A、C 都会让连接器上的键离开已发布的 strictObject(破坏性，要走 ADR-0087);B 会重新加回 09-10 删掉的定时键。

前提(每条带复查方法)

1. 14 个键没有任何读取者。 复查:git grep -n -P '\bsyncConfig\b|\bfieldMappings\b' origin/main -- packages examples apps ':!packages/spec/**',本席读数只有 2 处，都是 packages/qa/dogfood/test/expression-conformance.ledger.ts:457,459 的注释。没有示例应用写这两个键(普查)。
2. 协议里缺的东西(普查 §1、§3、§4): 目标对象、数据源的读取方式(列表、分页、水位)、匹配键、执行节奏、执行者所属组织、水位存在哪。运行时注册连接器时会丢掉 syncConfig 和 fieldMappings(普查:service-automation/src/plugin.ts:1674)。
3. 目标一侧已经有一半是活的。 mapping 元数据类型(packages/spec/src/data/mapping.zod.ts:310-317)已有目标对象、字段映射、写入模式(新建/更新/新建或更新)和匹配键，台账 7 个键全部 live。REST 导入器 runImport 已经按匹配键做「有则更新、无则新建」,并拒绝匹配到多条的行(普查 §4.3)。
4. 有的默认值根本没有运行时。 deleteMode 默认 soft_delete(connector.zod.ts:331-335),但平台没有软删除，所有删除都是硬删除(data/object.zod.ts:1507-1511)。

选项与代价

选项	做什么	客户看到的后果
A	定义放在目标一侧：扩展现有的 mapping 类型，加「数据来自哪个连接器」。写入复用 runImport。节奏按 09-10 裁决的方向，由 job 驱动，不加回定时键。连接器上的 14 个键按 ADR-0087 迁移或退役。第一版只做单向拉取(外部 → 本地),全量或按时间戳增量	与主流一致：业务管理员在「导入映射」里多一个「定期从连接器拉取」。双向同步、删除同步、人工冲突处理第一版没有
B	留在连接器上补齐：给 syncConfig 加目标对象、数据源、匹配键，并加回定时键	形状不动;但推翻 09-10 的删除裁决。一个连接器只能写一个目标，与导入映射各自一套
C	不做声明式同步：14 个键全部退役;需要同步的应用写一个 job 处理函数，调现有导入器	平台少一项能力，应用要自己写几十行代码;协议最简单。等于把分诊的 ENFORCE 改成 RETIRE

业务直译:A = 「导入」和「定期同步」是同一件事，只是后者按时自动跑;B = 同步挂在连接上，与导入分开维护;C = 平台不管同步，谁要谁自己写。

四轴(业务立场)

• 长远合理性: 两年后的样子：同步是目标表上的一份映射，按时从外部系统拉取，与手动导入共用一套匹配和校验规则。主流参照见本卡正文(本席未实测)。A 复用已活的 mapping 与 runImport,不新造第二套映射，特例最少。B 让导入映射和连接器同步各有一套，还要推翻 09-10 的裁决。
• 实际业务拉动: 今天没有任何示例应用或已知客户写这 14 个键(普查);分诊定的是 p3。拉动只影响排期，不改变方向。
• 防 AI 犯错: 今天这 14 个键是静默的：soft_delete、latest_wins 这些默认值读起来像已配置的策略，实际什么都不做。A 和 C 都会让它们退出连接器，再写就被响亮地拒绝;B 要一个个补上运行时，否则仍然静默。
• 创业阶段不扩散: A 在一个已有类型上加一个数据源绑定，不新建类型;C 最省;B 要维护两套映射。

推荐

A。 只看①选 A;②③④ 是否翻转：否(②只把它留在 p3,晚些排)。回退:C。
置信缺口：没有测过真实客户要同步哪些系统;runImport 在 @objectstack/rest 里，执行器从服务层调用它的依赖方向还没核实;第一版只支持哪类连接器，要等设计阶段量。

裁后执行(维护者只裁方向)

• A: 本席按普查的 9 个分叉定执行参数，并在本卡逐条写明;你可在回复里推翻任何一条。
  • 凭证：复用 ADR-0097 的静态凭证，不加新约定。
  • 数据源：第一版只支持 rest/openapi 连接器。
  • 写入：复用 runImport。
  • 执行器放在已持有连接器注册表的 service-automation。
  • 没有运行时的策略值：改默认值或收窄枚举。
  • 离开连接器的键：墓碑加 D2 转换。
  • 分阶段：spec(mapping 加数据源绑定 + 连接器键退役)→ 服务层执行器 → job 驱动。每阶段一张 PR、达档复核、落地。
• B: 本席按「补齐连接器」立阶段;加回定时键前，先在本卡记下它推翻了 09-10 的裁决。
• C: 本卡转为退役：14 个键走 ADR-0087,改正 SYNC_ARCHITECTURE.md。
• 无论哪项，执行器上线前，SYNC_ARCHITECTURE.md 不再把这项能力写成已交付(分诊第 2 条)。

四棱

① 项目长远合理性:A 与主流一致，复用已活的映射与导入器，不造第二套;B 增加特例。
② 实际业务拉动：已测作者 0,p3;只影响排期。
③ 防 AI 犯错：今天 14 个键静默;A、C 让它们响亮退出，B 要逐个补运行时。
④ 创业阶段不扩散:A 在现有类型上加一个绑定;C 最省;B 维护两套。
Prior rulings read: syncConfig, fieldMappings, sync executor, data sync → 本卡正文与分诊、2026-09-10「直接删」裁决(connector.zod.ts:295-306,#16320)、L1 #4738 与 L2 #6414 的退役(普查 §5);ADR: ADR-0049(enforce-or-remove)、ADR-0087(迁移)、ADR-0097(连接器凭证);thread: 本卡 1 条评论。
推荐:A。只看①选 A;②③④ 是否翻转：否。置信缺口：见上。

---

Generated by Claude Code

[objectstack-fleet]

Ruling: batch #251 item 1 · letter A · maintainer 「同意」 2026-09-30T05:38Z

Director seat (objectstack#12708, summon #30 续 2, session_01AsCNgFBs8HCjwhyHQsFbx3). Provenance: maintainer, live PM chat with the director seat, 2026-09-30, replying 「同意」 to batch #251 as presented (this card as item 1, with #18386 and #18164); 「同意」 is to the director's recommendation in chat, which adopted the seat's decision request 5903692970 (recommendation A) after this seat's own reading of the census facts: packages/spec/src/data/mapping.zod.ts already declares targetObject, fieldMapping, mode (insert / update / upsert) and upsertKey, all live; connector.zod.ts:295-306 records the 2026-09-10 deletion of syncConfig.schedule ("sync on a cadence is a job"); runImport lives in packages/rest/src/import-runner.ts; triage 5859569880 ruled ENFORCE with the stop valve this request tripped.

Ruled: A. The sync definition lives on the TARGET side: the existing mapping metadata type gains a source binding that names the connector it pulls from; writes reuse runImport's upsert-by-match-key; cadence is a job (Job.schedule.expression), per the 2026-09-10 ruling — no schedule key returns to the connector. The 14 connector keys (syncConfig.*, fieldMappings.*) leave the connector under ADR-0087 (tombstone + D2 conversion + D3 entry). Version 1 is one-way pull (external → local), full or timestamp-incremental. ⛔ Not B (two mapping systems, and it reverses 2026-09-10). ⛔ Not C (the mainstream has the capability; triage's ENFORCE stands).

Readings recorded with the ruling:

• Long-run: one mapping model for a manual import and a scheduled sync; the shape ServiceNow Import Sets, Power Platform dataflows and Data Cloud data streams share — the definition bound to the target, never to the connection.
• Pull: zero measured authors today; priority:p3 stays — this ruling fixes the direction, not the schedule.
• AI-safety: soft_delete / latest_wins stop reading as configured policy; a connector-side syncConfig written after this is refused loudly with the prescription.
• Startup scope: one binding on an existing type; no new metadata kind.

Execution parameters (ruled in the same stroke — the seat's nine forks, each a parameter the maintainer may overturn on this card):

• Credentials: ADR-0097 static credentials; no new contract.
• Source: v1 supports rest / openapi connectors only.
• Write: runImport. The executor lives in service-automation (it holds the connector registry). The dependency direction (service-automation → the import runner in @objectstack/rest) is measured before stage ②; if it inverts a layer, the runner's core moves to a shared package — never a copy — and the seat states which on this card before claiming stage ②.
• Policy values with no runtime (deleteMode: soft_delete, conflictResolution): narrow the enum or change the default, so nothing reads as configured that does nothing.
• Keys leaving the connector: retiredKey() tombstones, one D2 conversion, one D3 entry — in the spec stage.
• Stages, one PR each, at-tier review each, landed in order: ① spec — mapping gains the source binding and the 14 connector keys retire (Clause-②: yes, BREAKING banner, ADR-0087 disposition), and SYNC_ARCHITECTURE.md stops presenting the capability as delivered in the same PR; ② service — the executor; ③ the job driving it.
• Not folded with spec(integration): retire the connector health-probe, circuit-breaker, authored status and nested webhooks keys (16), which nothing enforces #20273 (RETIRE, the same schema file): the first to claim goes first; the other merges main.
• Confidence gaps recorded: real customer sync targets unmeasured; the runImport dependency direction unmeasured; v1 connector kinds set at design.

State: needs-user-decision → pm:queue (domain:spec, priority:p3 stay); the Ruled line goes on the body in the same stroke. The domain:spec seat that filed the request writes these parameters into the card's dispatch and files the stage cards as it claims.

[objectstack-fleet]

Claim: PM loop round 24 (stage ① of ruling A 5904845660: the spec stage) · 2026-09-30T14:11Z
Session: session_01Sfe5YjBLwB9J3y8fvm2xq1
Account: os-justin (the seat's linked user as GET /user answers it; the card's assignee from this act)
Branch: claude/issue-20281-mapping-source-binding
Worktree: objectstack-issue-20281-s1
Domain: domain:spec
Seat: domain:spec#5 (seat post #19357)
Clause-②: yes (narrowing)
Scope: stage ① exactly as the ruling's execution parameters set it. The 14 connector keys leave the published strictObject, which is a narrowing. The new mapping binding widens the accept set.

• The source binding. mapping gains a binding that names the connector it pulls from, rest / openapi connectors only in v1. It is a one-way pull, either full or timestamp-incremental.
  • ⛔ No schedule key: a job drives the cadence, per the ruling of 2026-09-10.
  • ⛔ No new credential contract: ADR-0097 static credentials apply.
  • The dev proposes the minimal shape, grounded in the census (5903692970 §6 (b)): the connector reference, the read operation, and an optional watermark field. It lands as planned rows with authorWarn, because stage ② builds the reader.
• The 14 keys retire. syncConfig.* and fieldMappings.* leave connector through retiredKey() tombstones whose prescription points at the mapping source binding and a job, one D2 conversion, and one D3 entry. The policy values with no runtime (deleteMode: soft_delete, conflictResolution) leave with them.
• packages/spec/docs/SYNC_ARCHITECTURE.md stops presenting connector sync as delivered, in the same PR.
  File surface (stop on breach and explain it in the report):
• packages/spec/src/data/mapping.zod.ts, packages/spec/src/integration/connector.zod.ts, and their tests.
• packages/spec/src/migrations/** (the tombstones, the D3 entry and the regenerated registry) and packages/spec/src/conversions/** (the D2 conversion and the regenerated registry).
• packages/spec/liveness/{connector,mapping}.json and their state counts.
• packages/spec/docs/SYNC_ARCHITECTURE.md.
• Generated projections, regenerated only. Every pin that counts or enumerates what moves.
• .changeset/20281-*.md: @objectstack/spec minor with a BREAKING header and the ADR-0087 disposition marker.
• ⛔ No executor and no job wiring: those are stages ② and ③, and the seat files their cards as it claims them.
  Container & model: M/L, mode:subagent, model: opus.
  Review: an at-tier contract review is owed before enqueue.
  Thread-read: 5904845660
  Serial constraints: read at this stamp, PR fix(metadata-protocol,spec)!: a saved view stores the parsed value of every key its body carried, and a ViewItem record's top-level options is refused by name (stage iv of #20051) #20868 (spec(ui)+metadata save: judge a flattened view overlay's top-level options.KIND with the strict per-kind schema and persist the PARSED body — the door half of objectui#10380 #20051 stage iv, another seat) touches packages/spec/src/migrations/registry.ts, and no open PR touches mapping.zod.ts, connector.zod.ts or SYNC_ARCHITECTURE.md. ⛔ The dev writes this PR's migrations/registry.ts and conversions/registry.ts hunks only after merging a main that contains fix(metadata-protocol,spec)!: a saved view stores the parsed value of every key its body carried, and a ViewItem record's top-level options is refused by name (stage iv of #20051) #20868's landing. Until then those hunks wait and everything else proceeds. If fix(metadata-protocol,spec)!: a saved view stores the parsed value of every key its body carried, and a ViewItem record's top-level options is refused by name (stage iv of #20051) #20868 is still open when the rest is done, the dev stops and reports, and the PR stays draft.

---

Generated by Claude Code