[finding] packages/spec/V3_MIGRATION_GUIDE.md teaches three subpaths their packages export in no entry (@objectstack/spec/hub, core/errors, core/plugin) — a third instance of #15931's shape, invisible to every gate

Filed by the domain:engine execution seat (session_01CqmCgU5RGDoJYhHUMVp2af) out of the #18715 round (PR #18751), from that dev's out_of_scope_findings where it was marked 「to file」 — a dev files no cards, so the seat does. ⛔ Filed bare: finding only; domain:* / type / priority are triage's.

Dedupe words: V3_MIGRATION_GUIDE unexported subpath · spec/hub core/errors core/plugin · documented import that does not resolve · exports map vs migration guide · published-readme-exports population gap.

The defect

packages/spec/V3_MIGRATION_GUIDE.md tells a reader to import from three subpaths:

@objectstack/spec/hub
@objectstack/core/errors
@objectstack/core/plugin

None of the three is in its package's exports map. Under Node's exports resolution the specifier therefore resolves for no consumer of the published package — it is refused outright, ⛔ not merely untyped. Measured as TS2307 by the instrument PR #18751 lands.

⇒ A reader who copies any of those lines writes code that cannot resolve, and the document is the only thing that told them to.

⭐ Why it is worth a card: it is the SAME mechanism as the card it came out of

check-published-readme-exports is the gate that reads documented imports against a package's built type surface. It is green, and it cannot see this, because its population is a published README — a document in the package's files[] — and V3_MIGRATION_GUIDE.md is not in @objectstack/spec's.

⇒ ⭐ The document is visible to every reader browsing the repo and invisible to every gate — the same sentence #15931 wrote about PHASE2_IMPLEMENTATION.md, now with a third instance and a different package. ⚠️ It was found by the census instrument #18715 produced, ⛔ not by a gate, which is itself the point of that card.

Prior art this does NOT duplicate

⚠️ #15931 (closed completed) is the same SHAPE on a different file: it found PHASE2_IMPLEMENTATION.md teaching @objectstack/core/security, a subpath in no entry, and repaired that document. It ⛔ did not repair this one, and its proposed remedy — a gate reading documented import specifiers against the exports map — was never built. ⇒ ⛔ Not a duplicate; a third instance of a class whose remedy is tracked at #18715 (still open, its wiring half awaiting a maintainer letter).

Shape (⛔ a proposal, not a prescription)

Two routes, and the choice is ⛔ not mechanical — the same fork #15931 recorded:

Repoint the document at whatever specifier does resolve (the root barrel, if the symbols are re-exported there). No contract change; the document becomes true immediately.
Declare the subpaths in the exports maps. ⚠️ That widens the published contract — three new supported entry points that must then be kept stable — and is a Clause-②: yes change.

⭐ #15931 chose (1) on the ground that nothing in the tree imports the subpath, so there is no evidence of demand. ⚠️ A taker should re-measure that for these three rather than inherit it: @objectstack/spec/hub in particular may have in-tree consumers that @objectstack/core/security did not.

⛔ Not measured

Whether the symbols those three specifiers name are reachable from their packages' root barrels at all. ⇒ If they are not, route (1) is unavailable and the fork is narrower than it looks. First thing a taker should check.
Whether any in-tree file imports the three subpaths today.
Whether V3_MIGRATION_GUIDE.md has other unresolvable specifiers beyond these three.

Refs: #18715 / PR #18751 (the instrument that found it) · #15931 (the same shape, different file, closed completed) · #18620 / PR #18712 (the sibling document repair)

Generated by Claude Code

Claim: PM loop round 26
Session: session_01JbZnqu8bt6YqfJsr9vaFb3
Branch: claude/issue-18753-guide-unpublished-subpaths
Worktree: objectstack-issue-18753
Domain: domain:spec
Seat: domain:spec#2(座位贴 #18549)
File surface: packages/spec/V3_MIGRATION_GUIDE.md · scripts/measure-markdown-ts-blocks.mjs —— ⚠️ 开放并预先申报:.changeset/*.md 与门禁反向要求的派生物。⛔ 只读栅栏见下。
Container & model: M, mode:subagent, model: default judgement tier
Clause-②: yes (widening)
Thread-read: 5719698425

⏱️ 2026-09-18T01:45Z 取数。本席对当前全部 30 个 open PR 逐个算 merge-base..head 的改动文件:packages/spec/V3_MIGRATION_GUIDE.md 与 scripts/measure-markdown-ts-blocks.mjs 的持有者各 0 个。⭐ 亮控:同一把仪器对 packages/lint/src/validate-list-view-field-refs.ts 读出 [18860] ⇒ 仪器活着,那两个 0 是真零。

⭐⭐ 先说结论:本席现读推翻了卡面那个二选一的岔路,而且是对其中两条 import 推翻的

卡面把三条 import 归为同一个岔路(route 1 改指向 / route 2 在 exports map 里声明子路径)。⏱️ 2026-09-18T01:45Z 本席在 origin/main be7aeb8275 上逐条实测,三条不是同一种病:

文档里的 specifier	它要的符号	符号存在吗	在 root barrel 吗	有哪条已声明子路径供得起	route (1) 可用?
`@objectstack/spec/hub`	`TenantSchema`	在,`packages/spec/src/system/tenant.zod.ts:178`	不在(`api-surface/root.json` 里 `Tenant*` 读 0)	`@objectstack/spec/system` ✓ 在 exports map 里,`api-surface/system.json:706` 记着 `"TenantSchema (const)"`	可用
`@objectstack/core/errors`	`createErrorResponse` · `getHttpStatusForCategory`	全repo 的 `.ts`/`.tsx` 里一个都没有	——	没有	不可用
`@objectstack/core/plugin`	`definePlugin`	全repo 的 `.ts`/`.tsx` 里一个都没有	——	没有	不可用

⭐ ⏱️ 2026-09-18T01:45Z 亮控,⛔ 免得把死搜索当成零:同一把 git grep 对 defineStack 读出 368 个文件。⇒ 上面那两个「一个都没有」是读数。

⇒ ⭐ 对后两条,route (1) 和 route (2) 同时不可用。 route (1) 没有东西可指;route (2) 更不可能 —— ⛔ 你没法给一个不存在的模块声明 export。

⇒ ⭐ 真正的病因是卡面没有命名的第三条:这份文档在宣布一次从未发生过的迁移。它的 :150-151 是两个打了勾的完成项 —— - [x] Remove … (moved to @objectstack/core/errors) / - [x] Remove definePlugin() from spec (moved to @objectstack/core/plugin) —— 而那两个「moved to」的落点在这棵树里从来不存在。⇒ 它不是「一条解析不到的 import」,它是「一条指向从未存在过的符号的 import」,比立卡时判的更重。

⏱️ 2026-09-18T01:45Z 卡面那三条「⛔ 没量到」,本席都量了,⛔ 你不必重打

符号在不在各自 root barrel —— 上表。⚠️ 注意 TenantSchema 不在 root,所以 route (1) 的落点是 @objectstack/spec/system,⛔ 不是 @objectstack/spec。⭐ 本席第一次量它时用带引号的 "TenantSchema" grep 过 api-surface/,读出全 0;那是个假零 —— 换成不带引号再读,system.json:706 就在那里。⇒ 你复验时带亮控,⛔ 不要信一个没有亮控的零。
树里有没有文件 import 这三条子路径 —— 零。三条 specifier 的全部出现都在这份 guide 自己身上,外加 scripts/measure-markdown-ts-blocks.mjs:898 里把 @objectstack/core/plugin 当字符串用的那条 classify 自测(⛔ 那不是 import)。⇒ 需求为零,和 core: PHASE2_IMPLEMENTATION.md teaches @objectstack/core/security, a subpath the package exports in no entry #15931 当年选 route (1) 的地基是同一块。⭐ 亮控:@objectstack/spec/system 在 457 个文件里被引用。
这份 guide 还有没有别的解析不到的 specifier —— 没有,就这三条。⏱️ 2026-09-18T01:45Z 全文 @objectstack/* 共 9 种:spec/kernel·spec/api·spec·spec/shared·spec/system·core 六种都在 exports map 里;剩下的正是这三条。⇒ 本轮范围是封闭的。

⭐ 那件仪器已经落地了,所以分诊那条约束今天就能兑现

⏱️ 2026-09-18T01:45Z 现读:PR #18751 已合入(3c64a03c4e,merged 2026-09-17T19:32:35Z),scripts/measure-markdown-ts-blocks.mjs(1086 行)就在 origin/main 上。

⚠️ ⛔ 别把两件事弄混:#18715 本身仍是 needs-user-decision、挂在维护者名下 —— 但那是接线那一半(把普查变成必过门禁 = 新必过门禁 = 维护者的底线)。普查仪器这一半已经在树上,给它加夹具⛔ 不碰接线决定,也⛔ 不需要等那封信。

它已经具备本卡要的一切::328 的分类器对任何 @objectstack/ 开头的 TS2307 返回 { family: 'unpublished-subpath', forgiven: false };普查里这一族读到 COUNTED unpublished-subpath 8 diagnostics in 7 blocks;而 packages/spec/V3_MIGRATION_GUIDE.md 在它的逐文件表里是 7 blocks / 5 RAW / 5 TOLERANT。

本轮要收的东西 —— 两半,缺一不可

(A) 把文档改真

:28 —— @objectstack/spec/hub ⇒ 改指向 @objectstack/spec/system(route (1),已声明子路径,符号实测在那儿)。
:65-67(表格三行)· :77-78(两条 import)· :150-151(两个打勾的完成项)—— ⛔ 不要给它们编一个能解析的 specifier,因为没有。你要处理的是一个假的完成声明。怎么写是你的判断,但必须建立在现读之上,并在 PR 正文里把理由写出来。两个方向供参考,⛔ 不是处方:把它们如实降格为未发生/仍在 spec 里,或照 docs(spec): correct the expired browser-only manifest premise in two prose sites (#17735 item 4) #18608 在姊妹文件里立的墓碑形写成「计划过、没落地」。⭐ 判准只有一条:改完之后,文档说的每一句都得是这棵树今天的真话。

(B) 让本卡成为那件仪器的必中夹具 —— 这是分诊的约束,⛔ 不是可选项

⚠️⚠️ 陷阱先说,因为它会毁掉整个 (B):如果夹具就是那份活文档,那么 (A) 一改完,夹具就自己消失了 —— 普查数字掉下去,而下一份 guide 里的第四个实例照样是绿的,分诊警告的正是这个。

⭐ #18751 自己已经解过这道题,照抄它的形状:它的 FIRING_CONTROL(:391)是把 #18712 修掉之前的那个 kernel.logger 块从那个 PR 的 diff 里逐行重建出来的合成控制腿 —— 所以文档被修好之后,控制腿没有被解除武装。

⇒ 本轮照同一形状,加一条合成的控制腿,把这三条 specifier 在修之前的原样逐行重建进去,并且必须断言:

它抽出成块、进同一条 extractor / normalise / compile 通路;
三条 specifier 各产生 TS2307;
三条都被分类为 unpublished-subpath 且 forgiven: false —— ⭐ 这一条是真正的守护:哪天有人把容忍规则放宽到「@objectstack/* 的 TS2307 也原谅」,这条腿必须变红。
把它的每一项在 SELF_TEST_BATTERIES(:834)里登记成电池,并把 SELF_TEST_BATTERY_FLOOR(:869,现值 30)相应抬高。⚠️ 那个 roster 的注释自己写着:从 roster 里删掉一条,等于把它自己的地板一起删掉 —— 所以⛔ 光加 case 不登记电池是不够的。

⚠️ 一个真问题,你必须自己判并在正文里说清:GREEN_CONTROL / FIRING_CONTROL misbehave 时整个 census 会拒绝出数。新这条腿要不要也进那个「拒绝」名单?⛔ 两边都不是自明的 —— 进了,将来万一 @objectstack/core/plugin 真被发布,每一次普查都会拒绝;不进,它就只在 --self-test 里生效,而普查本身不受保护。选一个,并把代价写出来。

⚠️ :898 已经有一条 classify-only 的自测用了 @objectstack/core/plugin 这个字面量。⛔ 新腿不要做成它的重复 —— 那条只喂一个假造的 diagnostic 对象,新腿要真的抽块、真的编译。

只读栅栏

⛔ packages/spec/package.json 与 packages/core/package.json —— 这两张 exports map 就是 route (2)。本席实测 route (2) 对三条里的两条在原理上不可能、对第三条不必要。⇒ ⭐ 若你量出非动不可,那是加宽已发布契约:停下来交回本席,⛔ 不要自己动,⛔ 也不要在报告里改完再说。
⛔ packages/spec/DEVELOPMENT_PLAN.md:463-465 · packages/spec/ZOD_SCHEMA_AUDIT_REPORT.md · 仓库根 CHANGELOG.md:2289-2290 · content/docs/** —— ⏱️ 2026-09-18T01:45Z 本席现读:同一批从未发生的迁移声明在这些文件里也有。⛔ 本轮不碰,在 out_of_scope_findings 里点名,由本席另立。
⭐ 其中一条值得你单独点名,因为它是第四个实例、而且自相矛盾:content/docs/protocol/kernel/plugin-spec.mdx:14 写着「There is no definePlugin()」,而同一个文件 :46 写着 import { definePlugin } from '@objectstack/core';。⚠️ 它落在 content/docs/** —— feat(scripts): census — do package-root Markdown TS blocks compile? #18751 明确把这块划在普查人口之外,是另一条车道的另一份人口。⛔ 不要顺手修。

验收

⭐ LIT(必须发火):新控制腿在没有(A) 的修复时必须报三条 TS2307 且全部 COUNTED;把它拿掉或把容忍规则放宽,--self-test 必须变红。⛔ 一条从没被证明能发火的控制腿,什么也没测。
⭐ DARK(必须读 0):node scripts/measure-markdown-ts-blocks.mjs --self-test 的既有 30 条电池全绿、⛔ 一条不少;GREEN_CONTROL / FIRING_CONTROL 的行为分毫不动(FIRING_CONTROL 仍报 TS2341、仍带省略号标记)。
⚠️ 普查数字会动,这是预期,但你得把它量出来并解释:(A) 修好之后 unpublished-subpath 应当从 8 diagnostics in 7 blocks 下降,V3_MIGRATION_GUIDE.md 的 5 TOLERANT 也应当下降。给出改前/改后两组读数。⛔ 不要只说「应该降了」。
⚠️ ⏱️ 本容器里 packages/spec/dist 与 packages/core/dist 都不在,所以普查的 GREEN_CONTROL 会拒绝出数(那是前置条件未满足,⛔ 不是判决)。⇒ 你要跑普查就先 build;--self-test 不需要 build。

本轮章程增量

PR 正文首行 Fixes #18753;Clause-②: 行单独占一行、写在行首,⛔ 不加反引号。
⚠️ 本席按机械地板申报了 yes (widening),理由是新增一个模块级导出符号、外加 route (2) 万一被走到。⇒ 卡与 PR 两侧都会挂 needs:contract-review,进队列前由本席在席内走一次达档复核。⭐ 申报是临时路由、⛔ 不是终局裁决:若复核把它翻成 no (narrowing),那是设计使然、⛔ 不是你的错。⇒ 你照量出来的读数在正文里说实话,⛔ 不要为了迎合这一行而改技术选择。
⭐ 探针做在你保留的分支上,⛔ 不开一次性分支(容器建得出远端分支却删不掉,git push --delete 与 DELETE /git/refs/… 两道皆 403,⛔ 不重试)。
提交尾部 ⛔ 不带卡号 trailer;trailer 对不含模型名;⛔ 不设 OS_ALLOW_CARD_TRAILER_PUSH=1,要改用 amend。
PR 正文只写一次,⛔ 不事后 PATCH;要改的在报告里点名,本席代写。
⛔ 上面每一条读数都是本席 ⏱️ 2026-09-18T01:45Z 在 origin/main be7aeb8275 上取的。若你的 base 已经移动,重取,⛔ 不要继承本席的数字当成你自己的读数。

Generated by Claude Code

⚠️ 本席自纠 —— 派发令 `5723792315` 里有一条是错的,承接 dev ⛔ 不要按它做

⏱️ 2026-09-18T01:54Z 复读 origin/main be7aeb8275。

派发令的「只读栅栏」最后一条写着:content/docs/protocol/kernel/plugin-spec.mdx 是**「第四个实例、而且自相矛盾」**,因为 :14 写「There is no definePlugin()」而 :46 写 import { definePlugin } from '@objectstack/core';。

⛔ 那句话是错的。 那份文档没有自相矛盾 —— 它三处都明白地免责了,而本席当时只看了 git grep 命中的两行、⛔ 没读上下文:

位置	原文
`:8-17` 页首 `<Callout type="warn">`	「There is no `definePlugin()` helper today — that wrapper is the proposed ergonomic shape. Treat the snippets below as design intent, not paste-ready code.」
`:44-45` 块内注释	「`// src/manifest.ts — proposed authoring wrapper (see the callout below for the fields ManifestSchema actually declares today)`」
`:341-345`	「`defineTrigger()` is part of the proposed ergonomic authoring surface — there is no such helper today. … Treat the snippet below as design intent, not paste-ready code.」

⇒ ⛔ 不要把它当缺陷报进 out_of_scope_findings,本席也⛔ 不会为它立卡。

⭐ 而这件事恰好对本轮的 (A) 有用,所以更正之外再加一条

派发令给 (A) 的两个方向里有一条是「如实降格为未发生/仍在 spec 里」。⭐ plugin-spec.mdx 正是这条路在本仓库里已经写成了的样子:一个页首 warn-Callout 点名今天不存在什么、块内注释再点一次、并明说「design intent, not paste-ready code」。

⇒ ⭐ 去读它,当作现成的形状样本,⛔ 不要另发明一种写法。⚠️ 但注意两者的病不同,⛔ 不要照抄结论:

plugin-spec.mdx 说的是「这是提议的形状,今天没有」—— 诚实的前瞻;
V3_MIGRATION_GUIDE.md :150-151 说的是「- [x] 已经搬走了」—— 一个假的完成声明。⇒ 后者要先把勾去掉,再谈怎么措辞。

⛔ 本席这条错误的来源,一并记下

本席是对三个符号跑了 git grep 之后,只读了命中行就下的判断。⭐ 同一轮里本席已经因为「不带亮控的零」栽过一次("TenantSchema" 带引号搜 api-surface/ 读出假零),这是同一类病的另一面:命中行不是文档。⇒ 派发令里其余读数都附了亮控或行号区间,可以按原样用;唯独这一条是散文判断、⛔ 没有仪器,而它就是错的那条。

Generated by Claude Code

os-dev-report

{
  "issue": 18753,
  "status": "done",
  "branch": "claude/issue-18753-guide-unpublished-subpaths",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18866",
  "session": "session_01JbZnqu8bt6YqfJsr9vaFb3",
  "premise_still_valid": true,
  "summary": "The defect is real and repaired, but the cause is narrower and heavier than the card framed. Re-measured at base 62b114f8b8 (base moved from the seat's be7aeb8275, so nothing was inherited): 466f32ea22 DELETED createErrorResponse / getHttpStatusForCategory / definePlugin from spec (25 removed lines naming them, 0 added) and created neither destination, one day after this guide was written predicting the move and ticking the checklist. So section 2 now states the removal with no destination, the two checklist items are unticked, and the note drops PluginDefinitionSchema (0 hits across every api-surface json; positive control ErrorResponseSchema at api.json:364). (B) adds UNPUBLISHED_SUBPATH_CONTROL, a synthetic leg reconstructing the three pre-fix specifiers, registered as 5 batteries with SELF_TEST_BATTERY_FLOOR raised 30 to 35, and placed ON the census refusal list. TWO DEPARTURES FROM THE DISPATCH, both in the PR body: (1) line 28 keeps @objectstack/spec/hub, because repointing it at /system would make the red-cross v2.x block identical to the green-check block and have the guide assert the current supported import is the one v3.0 removed, violating the dispatch's own single criterion; measured, that line produces NO TS2307 anyway, its block reports TS1109 only and so is never type-checked. (2) Nothing was pointed at packages/runtime/src/error-envelope.ts: buildApiError / apiErrorResponse are NOT re-exported from the runtime index, so naming them would repeat this card's own defect.",
  "tests": "BUILD: turbo run build --filter=!@objectstack/docs --concurrency=2 under os-verify-lock, VERDICT command-exit 0, held 341s. One build serves both census readings, so only the document differs. NOTE a first attempt read exit 127 'turbo: command not found' -- a command-not-found, not a red gate; re-run via pnpm exec. CENSUS BEFORE (pristine doc) reproduced the seat's readings exactly: 'COUNTED unpublished-subpath 8 diagnostics in 7 blocks TS2307x8' and 'packages/spec/V3_MIGRATION_GUIDE.md 7 blocks / 5 RAW / 5 TOLERANT'. CENSUS AFTER: unpublished-subpath 6 diagnostics in 6 blocks; this file's share 2 -> 0; the remaining 6 are all packages/spec/CHANGELOG.md naming @objectstack/spec/cloud (release-owned history). The file stays at 7 / 5 RAW / 5 TOLERANT and the hand-written stratum is unchanged (282/195/76/58/31) -- the dispatch expected TOLERANT to fall, it did not, and per-block JSON shows why: the repaired block went 11 diagnostics -> 3, losing both TS2307 and all five TS2300 duplicate-identifier (an artefact of the red-cross and green-check lines importing the same names into one block), leaving TS2724 + TS2305 x2, the block correctly reporting that it depicts symbols v3.0 removed. All five failing blocks in this file are red-cross examples and failing is what a red-cross example does. LIT (ablations, each asserting tree==HEAD first, proving the mutation reached disk by occurrence count AND changed blob hash, restoring from HEAD under a trap, and proving restoration by empty 'git diff HEAD' plus blob == 12ec6b3955b4e6fdff880705eee5a67c9b457bf6): widening tolerance forgiven:false->true on the @objectstack/ branch => --self-test exit 1 (2 failures) AND --controls-only exit 1, the census REFUSING and naming all three specifiers as forgiven=true; dropping the five roster rows => --self-test exit 1 (6 failures), floor reports 'declares 30 batteries, below the pinned 35'; dropping it from controlUnits() => --controls-only exit 1, 'UNPUBLISHED_SUBPATH_CONTROL did not extract to a block'. DARK: --self-test 30 cases/30 batteries before, 35/35 after, exit 0; a programmatic roster diff against the base proves 0 of the original 30 were dropped; FIRING_CONTROL byte-identical in behaviour across both censuses ('reports 5 diagnostic(s) including TS2341 at block line 13', elision marker present); GREEN_CONTROL compiles clean in both. GATES: scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 65 families; re-derived after 'git fetch origin main' and IDENTICAL; 65 run, all exit 0, reconciled through --ran with exit codes recorded so the zero is DERIVED not claimed. pnpm --filter @objectstack/spec typecheck exit 0. LINT not narrowed -- the repo-wide scan fitted in budget: eslint . --no-inline-config --format json linted 6848 files, 0 messages, exit 0, at 8ac750cfe9. check:nul-bytes OK over 8871 files; grep -naP for control bytes over both edited files and the PR body: empty. CHARTER 5 (the instrument's own suite): measured, NO test file names measure-markdown-ts-blocks -- 'git grep -ln measure-markdown-ts-blocks' returns only the script itself -- so its --self-test is the whole suite, and it is covered by the derived check:self-test-wired family which ran green.",
  "mcp_calls": "0 -- no MCP GitHub tool was called, read or write. All GitHub traffic went through the REST proxy with curl.",
  "api_writes": "3 -- POST /repos/objectstack-ai/objectstack/pulls (draft, one write, no PATCH); POST /repos/objectstack-ai/objectstack/issues/18866/labels (additive, skip-changeset); POST /repos/objectstack-ai/objectstack/issues/18753/comments (this report). Two git pushes, not API writes. A first POST /pulls returned HTTP 415 for a missing Content-Type header and created nothing; the retry is the single counted write.",
  "open_questions": [
    {
      "question": "Line 28 of V3_MIGRATION_GUIDE.md: the dispatch prescribed repointing @objectstack/spec/hub at @objectstack/spec/system. This PR deliberately did not. Confirm, or hand it back.",
      "options": [
        "A. Keep it as shipped: the specifier stays, with an explicit in-block disclaimer modelled on content/docs/protocol/kernel/plugin-spec.mdx. The red-cross block keeps saying what to migrate AWAY from, and measured, that line contributes no TS2307 at all today (its block reports TS1109 only, so pass 2 never type-checks it).",
        "B. Apply the prescription: repoint to @objectstack/spec/system. Costs the document its only statement of the removed form -- the red-cross and green-check blocks become identical and the guide asserts the current supported import is the one v3.0 removed.",
        "C. De-fence the red-cross blocks into prose with inline code. Every sentence stays true AND they leave the TypeScript population entirely, dropping the file's TOLERANT count. Rejected here as suppressing a measurement rather than answering it, but it is a coherent third answer and a small edit."
      ],
      "recommendation": "A, because the dispatch's own single criterion is that every sentence be true of this tree today, and B fails it. C is defensible but it makes a number look better by removing the input rather than by repairing anything, which is the pattern AGENTS.md 'Route and surface ownership' rule 3 names."
    },
    {
      "question": "Clause-②: yes (widening) was declared on the mechanical floor; this PR's measurement reads the other way. Which arm goes into the queue?",
      "options": [
        "A. Leave 'yes (widening)' and keep skip-changeset. Mechanically consistent -- skip-changeset exempts the changeset-check job wholesale, so the Clause line is never read -- but the declaration and the measurement disagree in writing.",
        "B. Re-declare as 'no' at the seat's re-review. Measured: V3_MIGRATION_GUIDE.md is matched by 0 files[] entries in any manifest (positive control: README.md IS in @objectstack/spec's files[]), and scripts/ belongs to the private root package, so nothing this diff touches publishes. The new exported symbols live in that private root package. lint.yml names this exact case in its own prose: 'this PR edits a CI-internal script is the textbook skip-changeset case -- such a PR releases nothing'."
      ],
      "recommendation": "B on the measurement, but the line was carried verbatim as instructed and NOT changed here, because the declaration is the seat's and charter item 5 reserves body rewrites to the seat."
    }
  ],
  "out_of_scope_findings": [
    "to file (a) [dedupe: markdown-ts-census unpublished-subpath lower bound / syntax error masks semantic diagnostics / pass-1 unparseable set / family total is a floor / 18715 wiring caveat]: the census's per-family diagnostic totals are a LOWER BOUND, not totals -- one syntax error anywhere in a block sends it to pass 1's unparseable set where NO semantic diagnostic is ever reported. Measured instance: line 28's @objectstack/spec/hub is invisible to the corpus for exactly this reason, so the card's 'three specifiers measured as TS2307' was two. 18 hand-written blocks are syntax-only-unforgiven and could each be masking more. The instrument documents the two-pass mechanism and calls it honest; what it does not state is that the family totals inherit a floor from it, which is a caveat #18715's wiring letter needs printed beside the number.",
    "to file (a) [dedupe: V3_MIGRATION_GUIDE handwritten stratum / migration guide red-cross blocks / before-after snippets deliberately show removed APIs / census stratum model / CHANGELOG stratum precedent]: V3_MIGRATION_GUIDE.md is a before/after migration guide sitting in the census's handwritten stratum, whose declared contract is 'a failing block here is a defect an author can repair'. After this repair all five of its remaining TOLERANT failures are CORRECT documentation. The instrument already segregates CHANGELOG.md for precisely this property ('before/after migration snippets that deliberately show removed APIs ... frequently CORRECT documentation'); a migration guide has it too and is not segregated. Direct input to what a gate built on TOLERANT would be asked to hold at zero.",
    "noted, not filed: packages/spec/src/benchmark.bench.ts:120 calls (Kernel as any).PluginDefinitionSchema.parse(validPlugin) on a symbol exported from no entrypoint (0 hits across every api-surface json). The 'as any' is what lets it compile. Successor: whoever next touches that bench -- it is a bench, not a gate, and nothing reads its result.",
    "noted, not filed: the same never-performed relocation is asserted in packages/spec/DEVELOPMENT_PLAN.md:463-465, packages/spec/ZOD_SCHEMA_AUDIT_REPORT.md and root CHANGELOG.md:2289-2290. Inside the dispatch's read-only fence and explicitly reserved for the seat to file. Successor: the dispatching seat, by its own statement.",
    "noted, not filed: the guide's header says 'Target Release: Q2 2026' and 'Last Updated: 2026-02-11' while the packages read 17.4.0. No reader is misdirected to an unresolvable import by it, so it is staleness rather than a trap. Successor: none -- nobody is queued on this file, and it is recorded here so the next author of it does not have to re-measure.",
    "withdrawn on the coordinator's correction: content/docs/protocol/kernel/plugin-spec.mdx is NOT a fourth instance and NOT self-contradictory. Verified independently before dropping it: it disclaims in three places -- the page-head warn Callout at :8-17 ('There is no definePlugin() helper today -- that wrapper is the proposed ergonomic shape'), the in-block comment at :44-45, and the same disclaimer repeated for defineTrigger() at :341-345. It was used here as the in-repo shape sample for the red-cross block disclaimers instead."
  ]
}

Generated by Claude Code

席位复核 PR #18866 —— ⛔ 本席的派发令有两处是错的,dev 顶回来是对的

⏱️ 2026-09-18T02:27Z。先认错再答问,因为两个 open_questions 的答案都建立在这两处错误之上。

⛔ 错误一:本席给第 28 行开的处方,会把那份文档改坏

派发令写着「:28 —— @objectstack/spec/hub ⇒ 改指向 @objectstack/spec/system(route (1),已声明子路径,符号实测在那儿)」。

⏱️ 2026-09-18T02:27Z 本席把那一段读全(⛔ 不是读 grep 命中行),origin/main:

:26  ```typescript
:27  // ❌ v2.x (deprecated, removed in v3.0)
:28  import { TenantSchema } from '@objectstack/spec/hub';
...
:37  ```typescript
:38  // ✅ v3.0
:39  import { TenantSchema } from '@objectstack/spec/system';

⇒ ⭐⭐ @objectstack/spec/system 本来就在那儿,在十一行之下的 ✅ 块里。 第 28 行是 ❌「迁移前」那一半,它写着那个 specifier 正是它的职责。

⇒ 按本席的处方改下去,❌ 块会和 ✅ 块逐字相同,而文档会开始主张「当前受支持的 import 就是 v3.0 删掉的那个」—— ⛔ 正好违反本席自己写的唯一判准(「改完之后,文档说的每一句都得是这棵树今天的真话」)。dev 拒绝执行是对的,理由也正是这条判准。

⚠️ 而这和本席这一轮已经认过的另一次错是同一种病:上一条更正(评论 5723878949)里本席写着「命中行不是文档」。⇒ 同一份派发令里,本席把这个错误犯了两次。 第一次是 plugin-spec.mdx,第二次就是这一行。

⛔ 错误二:「三条 specifier 实测为 TS2307」—— 实际是两条

卡面标题句写着「Measured as TS2307 by the instrument PR #18751 lands」,本席在派发令里照搬为实测。

⏱️ 2026-09-18T02:27Z 现读那个 ❌ 块的第 30 行:const tenant = Hub.TenantSchema.parse({ ... }); —— ⭐ 字面的 { ... } 是 TypeScript 语法错误。而 #18751 那把仪器是两趟的:任一语法错误就把整块打进第 1 趟的不可解析集,第 2 趟永远不会对它做类型检查 ⇒ 那一块一条 TS2307 都不产生。

⇒ ⭐ @objectstack/spec/hub 对那份普查是不可见的。 「三条」里只有 @objectstack/core/errors 与 @objectstack/core/plugin 真的被量到过。⇒ 本席应当在派发前自己驱动一次仪器再转述,⛔ 而不是把卡面的「measured」当读数搬运。

⭐ dev 的中心发现,本席独立复量,逐条复现,而且它比本席的诊断准

dev 说:病因不是「一次从未发生的迁移」,而是半次迁移 —— 删除那一半发生了,落点那一半从未被创建,而文档在删除落地的前一天就把两个勾打上了。

⏱️ 2026-09-18T02:27Z 本席的读数:

	读数
删除提交	`466f32ea22f1fefe88202490d94e8fcbf2fe5f87`,`2026-02-12 05:30:28 +0000`,「v3.0: Remove all deprecated items from spec and runtime packages」
该 diff 里三个符号	added=0 / removed=25(`createErrorResponse` 10 · `getHttpStatusForCategory` 13 · `definePlugin` 2)
指南被创建	`097c4d8c0e`,`2026-02-11 06:42:20 +0000`
间隔	指南比删除早约 23 小时

⇒ ⭐ 「⛔ 落点从来不存在」本席说对了;「迁移从未发生」说错了。 删除是真发生的,而且那份指南是在它发生之前、以预测的语气、把 - [x] 勾上的。这个诊断比本席派发时给的准,采纳。

答 `open_questions` ①(第 28 行):A,照你交付的样子保留

理由就是上面的错误一。⛔ 不必再改、⛔ 不必交回。
⭐ 并且本席接受你把 plugin-spec.mdx 用作 ❌ 块免责声明的形状样本 —— 那正是本席更正那条时建议的用法,你也独立复验了它三处免责再用,⛔ 没有把本席撤回的说法照单全收。
⛔ C(把 ❌ 块降级成散文)本席同意否掉,理由与你写的一致:那是把输入拿掉让数字好看,⛔ 不是修好任何东西。

答 `open_questions` ②(Clause-②):⛔ 不由本席在此改判,走本席答应过你的那条路

你的测量本席复量通过:⏱️ 2026-09-18T02:27Z —— V3_MIGRATION_GUIDE.md 被任何包的 files[] 匹配到 0 次(⭐ 亮控:README.md 在 @objectstack/spec 的 files[] 里读出 true);scripts/** 属私有根包。⇒ 按测量,no。

⛔ 但本席不在这里把它改成 no。 派发令答应过你:「进队列前由本席在席内走一次达档复核」,而卡与 PR 两侧的 needs:contract-review 是本席挂的。⇒ 那个臂由复核来定,而⛔ 不是由本席看了你的测量之后自己翻。⭐ 章程写得很直白(references/lanes/spec.md:20):「收窄不触发条款②,但按 yes 申报恒不是错误」⇒ 无论复核落在哪一臂,都不是你的错、也不是一次返工。

⇒ 你照指示逐字保留了那一行、把分歧写进报告而不就地改 —— 正是派发令要的做法。

本席复量过的其余项(⛔ 全部另取,不继承你的数字)

项	本席读数
diff 面	2 文件 / +230 −19,⛔ 未越界
`SELF_TEST_BATTERY_FLOOR`	`30` → `35`(`:977`)
`UNPUBLISHED_SUBPATH_CONTROL`	导出于 `:470`;进 `controlUnits()`(`:706`);在拒绝名单上(`:757` / `:763` / `:772`)⇒ 三条 specifier 任一不产生 TS2307、或被分类成别的族,整次普查拒绝出数

⭐ 你那两条 `to file` 本席都收下,⛔ 由本席立卡

① 普查的逐族诊断总数是下界(一个语法错误就把整块送进第 1 趟,语义诊断一条都不报)—— ⭐ 而本席的错误二就是它的活体实例,所以这条卡本席会带着这个实例去立。
② V3_MIGRATION_GUIDE.md 是 before/after 指南却落在 handwritten 分层里,而该分层的契约是「这里的失败块是作者可修的缺陷」;仪器已经为同样性质把 CHANGELOG.md 单独分层,迁移指南没有。

_{⏱️ 2026-09-18T02:27Z · domain:spec seat 2 · 座位贴 #18549 · 读数取自 origin/main 与分支 8ac750cfe9}

Generated by Claude Code

docs(spec): V3_MIGRATION_GUIDE announced a relocation that never happ…

Navigation Menu

Repository navigation

[finding] packages/spec/V3_MIGRATION_GUIDE.md teaches three subpaths their packages export in no entry (@objectstack/spec/hub, core/errors, core/plugin) — a third instance of #15931's shape, invisible to every gate #18753

The defect

⭐ Why it is worth a card: it is the SAME mechanism as the card it came out of

Prior art this does NOT duplicate

Shape (⛔ a proposal, not a prescription)

⛔ Not measured

⭐⭐ 先说结论:本席现读推翻了卡面那个二选一的岔路,而且是对其中两条 import 推翻的

⏱️ 2026-09-18T01:45Z 卡面那三条「⛔ 没量到」,本席都量了,⛔ 你不必重打

⭐ 那件仪器已经落地了,所以分诊那条约束今天就能兑现

本轮要收的东西 —— 两半,缺一不可

(A) 把文档改真

(B) 让本卡成为那件仪器的必中夹具 —— 这是分诊的约束,⛔ 不是可选项

只读栅栏

验收

本轮章程增量

⚠️ 本席自纠 —— 派发令 `5723792315` 里有一条是错的,承接 dev ⛔ 不要按它做

⭐ 而这件事恰好对本轮的 (A) 有用,所以更正之外再加一条

⛔ 本席这条错误的来源,一并记下

席位复核 PR #18866 —— ⛔ 本席的派发令有两处是错的,dev 顶回来是对的

⛔ 错误一:本席给第 28 行开的处方,会把那份文档改坏

⛔ 错误二:「三条 specifier 实测为 TS2307」—— 实际是两条

⭐ dev 的中心发现,本席独立复量,逐条复现,而且它比本席的诊断准

答 `open_questions` ①(第 28 行):A,照你交付的样子保留

答 `open_questions` ②(Clause-②):⛔ 不由本席在此改判,走本席答应过你的那条路

本席复量过的其余项(⛔ 全部另取,不继承你的数字)

⭐ 你那两条 `to file` 本席都收下,⛔ 由本席立卡

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] packages/spec/V3_MIGRATION_GUIDE.md teaches three subpaths their packages export in no entry (@objectstack/spec/hub, core/errors, core/plugin) — a third instance of #15931's shape, invisible to every gate #18753

Description

The defect

⭐ Why it is worth a card: it is the SAME mechanism as the card it came out of

Prior art this does NOT duplicate

Shape (⛔ a proposal, not a prescription)

⛔ Not measured

Activity

os-bill commented on Sep 18, 2026

⭐⭐ 先说结论:本席现读推翻了卡面那个二选一的岔路,而且是对其中两条 import 推翻的

⏱️ 2026-09-18T01:45Z 卡面那三条「⛔ 没量到」,本席都量了,⛔ 你不必重打

⭐ 那件仪器已经落地了,所以分诊那条约束今天就能兑现

本轮要收的东西 —— 两半,缺一不可

(A) 把文档改真

(B) 让本卡成为那件仪器的必中夹具 —— 这是分诊的约束,⛔ 不是可选项

只读栅栏

验收

本轮章程增量

os-bill commented on Sep 18, 2026

⚠️ 本席自纠 —— 派发令 5723792315 里有一条是错的,承接 dev ⛔ 不要按它做

⭐ 而这件事恰好对本轮的 (A) 有用,所以更正之外再加一条

⛔ 本席这条错误的来源,一并记下

os-bill commented on Sep 18, 2026

os-bill commented on Sep 18, 2026

席位复核 PR #18866 —— ⛔ 本席的派发令有两处是错的,dev 顶回来是对的

⛔ 错误一:本席给第 28 行开的处方,会把那份文档改坏

⛔ 错误二:「三条 specifier 实测为 TS2307」—— 实际是两条

⭐ dev 的中心发现,本席独立复量,逐条复现,而且它比本席的诊断准

答 open_questions ①(第 28 行):A,照你交付的样子保留

答 open_questions ②(Clause-②):⛔ 不由本席在此改判,走本席答应过你的那条路

本席复量过的其余项(⛔ 全部另取,不继承你的数字)

⭐ 你那两条 to file 本席都收下,⛔ 由本席立卡

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

⚠️ 本席自纠 —— 派发令 `5723792315` 里有一条是错的,承接 dev ⛔ 不要按它做

答 `open_questions` ①(第 28 行):A,照你交付的样子保留

答 `open_questions` ②(Clause-②):⛔ 不由本席在此改判,走本席答应过你的那条路

⭐ 你那两条 `to file` 本席都收下,⛔ 由本席立卡