[finding] `SYNC_ARCHITECTURE.md` 与 `flows.mdx` 把 `retryConfig` 教成上游限流的解法并列出它的默认值,而实测零消费者、`ConnectorProviderContext` 根本不携带它 —— 照文档做的作者拿到的是被静默忽略的配置(同一文件的 `rateLimitConfig` 半边已由 #5554 收过)

[os-bill]

⏱️ 本卡所有读数取自同一动作:2026-09-17T21:06Z。本席第一手复核过 dev 交上来的每个数。

一句话

packages/spec/docs/SYNC_ARCHITECTURE.md 与 content/docs/automation/flows.mdx 教作者把 retryConfig 当成上游限流的解法,并列出它的 retryableStatusCodes 默认值;而实测没有任何东西读它,连 ConnectorProviderContext 都不携带它。⇒ 照着文档做的作者,拿到的是被静默忽略的配置。

实测

⏱️ 下面这块读于 2026-09-17T21:06Z。

读取式探针(属性访问 = 真正的消费者),packages/spec 之外:
  .retryConfig            0
  .connectionTimeoutMs    0
  .requestTimeoutMs       0
⭐ LIT 对照:同一 schema 上确有消费者的兄弟键 `providerConfig`,同一判据下命中
   connector-mcp/src/mcp-provider.ts · connector-openapi/src/openapi-provider.ts
   · connector-rest/src/rest-provider.ts 等
⇒ 上面三个 0 是**读数**,⛔ 不是搜索没响。

宿主接缝(「留给宿主」这一支的判据):
  packages/spec/src/integration/connector-provider.ts:57
  ConnectorProviderContext = name,label,description,icon,type,providerConfig,auth,loadPackageFile
  ⇒ 三个键**一个都不在**其中 ⇒ provider 工厂**没有能力**兑现它们。

那两处教学文本:
  packages/spec/docs/SYNC_ARCHITECTURE.md:157 / :336 / :360
  content/docs/automation/flows.mdx:1588

⇒ 三分支判定:declared-but-unimplemented = 真;left-to-the-host = 假;pending-retirement = 未确立(ADR-0049 欠一次裁定)。

⭐ 同一个文件为同一个形已经收过一次

⏱️ 读于 2026-09-17T21:06Z。查重(MCP search_issues,含 closed):

• SYNC_ARCHITECTURE.md 的 L3 段仍在三处宣传 Rate Limiting,而 connector.rateLimitConfig 及其整个形状已在 #4911 退役 #5554(closed/completed):「SYNC_ARCHITECTURE.md 的 L3 段仍在三处宣传 Rate Limiting,而 connector.rateLimitConfig 及其整个形状已在 「v17」ConnectorRateLimitConfig 出站限流词表按 ADR-0049 摘除 —— 执行引擎不存在(#4686 拆向之二) #4911 退役」。
  ⇒ 同一个文件、同一个形、不同的键。那次收的是 rateLimitConfig,这次是 retryConfig 与它旁边的两个超时。
• ADR-0049:spec 两份 RateLimitConfig 全仓零 reader,真正在限流的是 runtime 自己的第三份形状 #4686 / 「v17」ConnectorRateLimitConfig 出站限流词表按 ADR-0049 摘除 —— 执行引擎不存在(#4686 拆向之二) #4911(均 closed):ADR-0049 下把 RateLimitConfig 摘除的那条线。
• spec/liveness: seed ledgers for the three PENDING_GOVERNANCE debts #18133 declared — connector, sharing_rule, analytics_cube #18582(closed/completed):把 connector 等三份 PENDING_GOVERNANCE 账本播下去的卡。

⚠️ 无 open 的 ADR-0049 决策卡在管 retryConfig/超时这一半 —— 本席逐条读过上面每一张的态。⇒ 没有可并入的既有条目,故另立。

为什么本卡与 #18614 分开

#18614 收的是 conversions/registry.ts 里那句**「they are live」的断言**(已由 PR #18789 收掉)。本卡收的是教学文本把作者引向一个死键——不同文件、不同危害(一个是错误记载,一个是把人引去写没用的配置)。⭐ dev 明确没有把它折进 #18614,理由是文件面外且卡面自陈不要求退休 —— 处置正确。

建议(⛔ 非裁定)

• A —— 只改文档:把这三个键的教学文本改成「已声明但当前无实现」,并指向 packages/spec/liveness/connector.json。⚠️ 最小,且不预判 ADR-0049。
• B —— 等 ADR-0049 对这三个键裁定(退休 / 实现 / 声明为宿主契约),再一次性改文档与声明。⚠️ 代价是在此期间作者继续被误导。
• C —— 先按 A 止血,并把 ADR-0049 的那次裁定单独立卡。

⭐ 本席倾向 C:止血便宜,而裁定不该被一条文档修复顺手代做。⛔ 由维护者定。

⚠️ 另记(⛔ 不另立卡):content/docs/references/integration/connector.mdx 也在宣传这三个键及其默认值,但它是从声明生成的(packages/spec/scripts/build-docs.ts),⇒ 声明或生成器一动它就跟着动,没有独立的缺陷。

出处

domain:spec seat 2(座位贴 #18549)复核 PR #18789 / 卡 #18614 时,dev 在 out_of_scope_findings 里交出来的,并明确请分诊先对 ADR-0049 的决策条目查重。⭐ 那次查重由本席完成(隔离子代理够不到 /search/*),结论是没有可并入的条目。

---

Generated by Claude Code

[os-bill]

Claim: PM loop round 32
Session: session_01JbZnqu8bt6YqfJsr9vaFb3
Branch: claude/issue-18794-retryconfig-declared-unimplemented
Worktree: objectstack-issue-18794
Domain: domain:spec
Seat: domain:spec#2(座位贴 #18549)
File surface: packages/spec/docs/SYNC_ARCHITECTURE.md —— 仅此一个。 ⚠️ 卡面点的第二个文件 content/docs/automation/flows.mdx 被本轮排除,理由见下。
Container & model: M, mode:subagent, model: default judgement tier
Clause-②: no
Thread-read: none

⏱️ 2026-09-18T08:56Z 取数,base origin/main = 106717c3aa。⛔ 以下每条本席本轮重取,base 移动就重取。

---

本轮走 A,⛔ 不走 B —— 并说清为什么这不是替维护者裁

卡面给了三条(A 只改文档止血 / B 等 ADR-0049 裁完再一次性改 / C = A + 另立裁定卡)。

⭐ A 改的是「今天实测为真的那句话」,⛔ 不预判任何裁定 —— 卡面自己写明 A「不预判 ADR-0049」。⇒ 把教学文本改成与实测一致,是事实更正,⛔ 不是契约取向。B 的代价是「在此期间作者继续被误导」。

⇒ 本轮做 A;C 的另一半(ADR-0049 那次裁定)由本席另立卡,⛔ 不要你顺手代做。

本席重取的读数(卡面读于 2026-09-17T21:06Z,行号已漂)

⏱️ 现读 origin/main 106717c3aa

教学站点(本轮可动面):
  packages/spec/docs/SYNC_ARCHITECTURE.md:157   「rate-limited upstream 的解法是 retryConfig」
  :301 / :307                                    retryConfig 块与 retryableStatusCodes 默认值
  :336                                           「retryConfig handles the 429 …」
  :360                                           决策表把它列为选 L3 的理由

⭐ 宿主接缝这一支(卡面判「left-to-the-host = 假」的依据),本席逐字复核:
  packages/spec/src/integration/connector-provider.ts:57  export interface ConnectorProviderContext {
    name · label · description? · icon? · type · providerConfig · auth? · loadPackageFile?
  ⇒ retryConfig / connectionTimeoutMs / requestTimeoutMs 三个键 **一个都不在其中**
  ⇒ provider 工厂没有能力兑现它们 ⇒ 「留给宿主」这条辩解不成立

⚠️ 卡面的三个「0 消费者」读数本席未重跑(.retryConfig / .connectionTimeoutMs / .requestTimeoutMs 的属性访问探针,卡面配了 providerConfig 作阳性对照)。⇒ 记作卡面的读数,⛔ 不是本席的。你自己重取,并且必须带同形阳性对照 —— 没有对照的 0 不是读数。

⛔ 只读栅栏

• ⛔⛔ content/docs/automation/flows.mdx —— ⏱️ 2026-09-18T08:56Z 实测被 open PR feat(spec,types,triggers)!: group runs package-authored scheduled work without a declaration, owning each run's writes per record #18420 持有。⭐ 本席读了那一处(现读 :1595,卡面写 :1588,已漂):它讲的是几种 retry 块之间计数口径的换算(maxAttempts 含首次尝试,故 maxAttempts: 3 等于这里的 maxRetries: 2),⛔ 不是「用 retryConfig 解决上游限流」那条处方。⇒ 本轮排除它不损伤卡的主张。在报告里点名它为后继项(待 feat(spec,types,triggers)!: group runs package-authored scheduled work without a declaration, owning each run's writes per record #18420 落地后收),⛔ 不要动它。
• ⛔ content/docs/references/integration/connector.mdx —— 卡面写明它是从声明生成的(packages/spec/scripts/build-docs.ts),声明或生成器一动它就跟着动,没有独立缺陷。⛔ 不手改生成物。
• ⛔ packages/spec/src/integration/** 的任何声明 —— 本轮不退休、不实现、不改 schema。那是 ADR-0049 的裁定,归维护者。撞到即停。
• ⛔ packages/spec/liveness/connector.json —— 可以指向它,⛔ 不改它。

验收

• ⭐ 主体腿:改后 SYNC_ARCHITECTURE.md 的每一处必须说清「已声明但当前无实现」,并指向 packages/spec/liveness/connector.json。⛔ 不许改成「已退休」或「宿主负责」—— 两者今天都不成立(后者本席已逐字证伪,见上)。
• ⭐ 必响对照(同一次运行):你的消费者探针必须同时给出 providerConfig(有消费者)的命中,与那三个键的 0。⛔ 单给 0 不算读数。⚠️ 用免折行判据(剥注释前缀 + 折叠含换行的空白)—— 本班已三次栽在按行探针上。
• ⭐ 决策表那一行(:360):它今天把 retryConfig 列为「选 L3 的理由」。改后必须不再把一个无实现的键当作选型理由,⛔ 但也不许顺手改动同一行里关于 #4911 的既有结论。
• DARK:packages/spec 的测试与 check:generated 改前改后不动(本轮只改一份非生成的 .md)。
• 门禁清单取 node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack 逐条跑,退出码先落盘再 --ran 对账。
• changeset:skip-changeset。⚠️ 本轮的停手线按「消费者能读到的内容是否变化」判,⛔ 不按字节 —— packages/spec 的 files[] 是否包含 docs/ 请你实测后在报告里给读数(⭐ 若它发布,那就不是 skip-changeset,停下来交回本席)。

本轮章程增量

1. PR 正文首行 Fixes #18794;Clause-②: no 行单独占一行、写在行首,⛔ 不加反引号。
2. ⚠️ 本席申报 no:只改散文,⛔ 不动任何声明、schema 或 accept set。量出不是这样就停下来交回本席。
3. ⭐ 探针做在你保留的分支上,⛔ 不开一次性分支(删除远端分支两道皆 403,⛔ 不重试)。
4. 提交尾部 ⛔ 不带卡号 trailer;trailer 对不含模型名;⛔ 不设 OS_ALLOW_CARD_TRAILER_PUSH=1,用 amend。
5. PR 正文只写一次,⛔ 不事后 PATCH;要改的在报告里点名,本席代写。
6. ⚠️ ⭐ 本席在本班已被 dev 正当地顶回来八次,其中四次是约束写得过于字面(三次载体是只读栅栏或停手线,最近一次:把「tarball 会动」写成停手线,而任何注释改动都会移动 sourcemap 的 mappings)。⇒ 本令任何一条你量出做不到、或与卡面/实测抵触,照实顶回来 —— 八次没有一次是 dev 错。
7. ⚠️ 卡上已有本席贴的 Claim: 评论。⛔ 不要再贴第二条。
8. ⛔ 本令读数取自 origin/main = 106717c3aa,⏱️ 2026-09-18T08:56Z 取。base 已移动就重取。

---

Generated by Claude Code

[os-bill]

os-dev-report

{
  "issue": 18794,
  "status": "done",
  "branch": "claude/issue-18794-retryconfig-declared-unimplemented",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18979",
  "session": "session_01JbZnqu8bt6YqfJsr9vaFb3",
  "premise_still_valid": true,
  "summary": "Option A only. Five passages in packages/spec/docs/SYNC_ARCHITECTURE.md (:157 blockquote, :299 example comment, :312 example timeouts, :336 Best Practices, :360 decision row) now say retryConfig and the two timeouts are declared but currently unimplemented, each pointing at packages/spec/liveness/connector.json; none says retired and none says the host's job. No declaration, schema or accept set is touched, so the ADR-0049 ruling is not prejudged, and the :360 row's existing #4911 sentence is left byte-identical. ONE WIDENING TO CONFIRM: the :360 row and the :157 blockquote both paired retryConfig with health.circuitBreaker, which the ledger also records dead and which my probe also measures at 0 read-shaped consumers outside packages/spec, so correcting only the retryConfig half would have left the row still presenting an unimplemented key as a selection criterion; both halves are corrected, citing the ledger's existing verdict rather than making a new one. Issue assignee was os-bill on arrival (PM's dispatch step 1); I did not write it.",
  "tests": "GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 48 commands at commit c4c09b853e; all 48 run with exit codes landed to disk BEFORE any pipe, then reconciled via --ran: '48 derived, 48 run, 0 NOT-MEASURED, 0 UNRUN' and 'a DERIVED zero - all 48 recorded an exit code and none of them is 3'. Four (check:dts-closure, check:dual-build-cjs-loads, check:lean-entry-closure, check:sourcemap-no-sources-content) first exited 3 = PREREQUISITE NOT MET (no dist), recorded as NOT MEASURED, not as red; after 'pnpm build' under the shared verify lock (OS_VERIFY_LOCK_SLOT=issue-18794-build, VERDICT command-exit 0, 73/73 tasks) all four re-ran exit 0. DARK LEG: pnpm --filter @objectstack/spec test = 488 files / 14182 tests all pass (lock VERDICT command-exit 0); pnpm --filter @objectstack/spec check:generated = 'All 15 generated artifacts are up to date'; pnpm --filter @objectstack/spec typecheck clean. DOC COMPILE PIN: packages/spec/src/integration/connector-author-shape.test.ts compiles this file's typescript fences verbatim and pins fence count 2 / elision sketches 0 - 14/14 pass with the edit in place (fence count re-measured on disk = 2). REVERSE VERIFICATION (direction stated before running: must turn RED on the block I edited): scripts/ablation-replace.mjs renamed the edited block's retryConfig to retryConfigBogus; mutation landing proved on disk (anchor 1 to 0, blob bd3c6b895c2a to c02cf5c9c5e3), gate went RED with TS2561 'Object literal may only specify known properties' on 'L3 example #0 must compile clean' (1 failed / 13 passed) - observed code is TS2561, the did-you-mean form of the excess-property error, not the TS2353 I predicted; restore proved independently of the tool's claim (blob back to bd3c6b895c2a = HEAD, git diff HEAD empty, git status clean). CONSUMER PROBE (fold-proof: comment leaders stripped, then all whitespace incl. newlines collapsed and glued to access punctuation; read-shaped access only; 8763 tracked files at tree 106717c3aa): .retryConfig 0 hits in any packages/ or examples/ file (2 total, both in the GENERATED content/docs/references/integration/connector.mdx); .connectionTimeoutMs 0; .requestTimeoutMs 0; FIRING CONTROL .providerConfig 15 hits across 9 files (connector-mcp/openapi/rest providers, service-automation plugin.ts, app-showcase tests) - same predicate, same run. SECOND CONTROL (reachability): bare-identifier census shows connectionTimeoutMs in 9 files and requestTimeoutMs in 8 files outside packages/spec, every one a WRITE of the literal into a def, never a read - so the zeros are 'no reader', not 'the probe never looked'. HOST SEAM re-read verbatim: connector-provider.ts:57 ConnectorProviderContext = name, label, description?, icon?, type, providerConfig, auth?, loadPackageFile? - none of the three keys, card's falsification CONFIRMED. CHANGESET MEASURED: npm pack --dry-run --json --ignore-scripts on packages/spec = 275 entries, docs/SYNC_ARCHITECTURE.md ABSENT, zero docs/ paths at all, positive controls liveness/connector.json and src/integration/connector.zod.ts both PRESENT; files[] carries no docs entry - so skip-changeset holds and the stop-line did not fire. LINT NARROWING DECLARED: (1) population from eslint's own config - every files: selector in eslint.config.mjs targets {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown selector or processor; (2) count from --format json - the changed file yields 'File ignored because no matching configuration was supplied.', 0 errors, so 0 files of the lint population are touched; (3) invariance - the file is outside the lint population entirely, so no untouched file's verdict can move. CONTROL CHARS: pnpm check:nul-bytes exit 0 plus a direct grep -naP scan of the changed file, exit 1 = none found (exit captured before any pipe).",
  "mcp_calls": "0 - no MCP GitHub tool was called, read or write; every GitHub touch went through the REST proxy with curl",
  "api_writes": "3 - POST /repos/objectstack-ai/objectstack/pulls (draft PR 18979); POST /repos/objectstack-ai/objectstack/issues/18979/labels (additive, skip-changeset); POST /repos/objectstack-ai/objectstack/issues/18794/comments (this report). Plus two git pushes (the empty-branch routing probe, then the commit) - no 403 on either. No PATCH of the PR body.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: packages/spec/src/integration/connector.zod.ts:44 carries the SAME sentence verbatim ('What L3 does declare for a rate-limited upstream is retryConfig - whose retryableStatusCodes default ... includes 429'), and content/docs/references/integration/connector.mdx:42 is that comment regenerated by packages/spec/scripts/build-docs.ts. Both sit behind this round's fence on packages/spec/src/integration/**, so after this PR the repo still teaches the falsehood in those two places from one source. Follow-up: correct that TSDoc block and regenerate (prose-only change plus gen:docs). Successor: whoever takes the ADR-0049 ruling card, since it lands in the same file. NOT filed by me - it is inside the fence and the PM owns the ruling card.",
    "noted, not filed: content/docs/automation/flows.mdx has drifted to :1595 (card said :1588) and is held by open PR #18420 (verified open, draft, 23 files incl. that path). Its passage is about reconciling retry-COUNT conventions between block kinds, not the rate-limiting prescription, so excluding it does not damage the card's thesis. Follow-up: sweep once PR #18420 lands. Successor: the seat that picks up that sweep.",
    "noted, not filed: SYNC_ARCHITECTURE.md:192 also names connectionTimeoutMs / requestTimeoutMs, but only to state that .default()-carrying keys are optional in the author shape - a true statement about z.input making no efficacy claim. Left alone deliberately; successor: none needed.",
    "noted, not filed: the Best Practices bullet 'Error Handling: Implement comprehensive retry logic with exponential backoff' is left as-is - under this change it reads correctly as advice to IMPLEMENT retry yourself. Successor: none needed.",
    "PLATFORM STATE I DID NOT SET (reported, not corrected): on creating PR 18979 the platform APPENDED a second attribution footer block - my rule-less session-URL footer survived verbatim and a '---' + session-URL footer was appended after it, exactly +90 bytes, stored body = sent body byte-identical as a prefix. This matches platform-readings line 349 for the create action. Per the round convention I wrote the body once and did NOT PATCH it; if the duplicate should be stripped, that is the PM's write (the PATCH recipe is platform-readings 345-347). Also: a 'documentation' label was added to PR 18979 by another actor after my additive label write; my comparative read-back shows nothing stripped (union(read, target) minus read-back = none), so I left it alone."
  ]
}

---

Generated by Claude Code