diff --git a/docs/architecture/rfcs/STATUS.md b/docs/architecture/rfcs/STATUS.md
index f67af4274c..e98c033cd8 100644
--- a/docs/architecture/rfcs/STATUS.md
+++ b/docs/architecture/rfcs/STATUS.md
@@ -54,7 +54,7 @@ appendix may keep dated history, but no dated log heading may precede it.
 | [RFC: Live Team Workspace v0](live-team-workspace-v0.md) | Accepted | none | — |
 | [RFC: Long-Horizon Harness Benchmark and Research Program v0](long-horizon-harness-benchmark-research-program-v0.md) | Accepted | none | — |
 | [RFC: Long-Running Agent Reliability Diagnostics and Governed Delivery v0](long-running-agent-reliability-diagnostics-governed-delivery-v0.md) | Accepted | none | — |
-| [LoopX Overall Roadmap v0: Product, Collaboration, Technology and Delivery](loopx-overall-roadmap-v0.md) | Accepted | none | [1 entry](ledger/loopx-overall-roadmap-v0/) |
+| [LoopX Overall Roadmap v0: Product, Collaboration, Technology and Delivery](loopx-overall-roadmap-v0.md) | Accepted | none | [2 entries](ledger/loopx-overall-roadmap-v0/) |
 | [Manager runtime profile v0](manager-runtime-profile-v0.md) | Accepted | none | — |
 | [RFC: Monorepo Distribution and Small-Kernel Runtime Composition (v0)](monorepo-distribution-split-v0.md) | Accepted | none | — |
 | [RFC: Obelisk Session Evidence Provider v0](obelisk-session-evidence-provider-v0.md) | Accepted | none | — |
diff --git a/docs/architecture/rfcs/STATUS.zh-CN.md b/docs/architecture/rfcs/STATUS.zh-CN.md
index 373aa3ef3c..2944938f69 100644
--- a/docs/architecture/rfcs/STATUS.zh-CN.md
+++ b/docs/architecture/rfcs/STATUS.zh-CN.md
@@ -51,7 +51,7 @@
 | [RFC：团队实时工作区 v0](live-team-workspace-v0.zh-CN.md) | 已接受 | 无 | — |
 | [RFC：长程 Harness Benchmark 与研究计划 v0](long-horizon-harness-benchmark-research-program-v0.zh-CN.md) | 已接受 | 无 | — |
 | [RFC：长程 Agent 可靠性诊断与治理交付 v0](long-running-agent-reliability-diagnostics-governed-delivery-v0.zh-CN.md) | 已接受 | 无 | — |
-| [LoopX 整体路线总纲 v0：产品、协作、技术与交付](loopx-overall-roadmap-v0.zh-CN.md) | 已接受 | 无 | [1 条](ledger/loopx-overall-roadmap-v0/) |
+| [LoopX 整体路线总纲 v0：产品、协作、技术与交付](loopx-overall-roadmap-v0.zh-CN.md) | 已接受 | 无 | [2 条](ledger/loopx-overall-roadmap-v0/) |
 | [Manager runtime profile v0 / 管家运行模式 v0](manager-runtime-profile-v0.zh-CN.md) | 已接受 | 无 | — |
 | [RFC：Monorepo 发行拆分与小内核运行时装配（v0）](monorepo-distribution-split-v0.zh-CN.md) | 已接受 | 无 | — |
 | [RFC：Obelisk Session Evidence Provider v0](obelisk-session-evidence-provider-v0.zh-CN.md) | 已接受 | 无 | — |
diff --git a/docs/architecture/rfcs/ledger/loopx-overall-roadmap-v0/2026-10-30-s2-s12-windows-startup-delivery.md b/docs/architecture/rfcs/ledger/loopx-overall-roadmap-v0/2026-10-30-s2-s12-windows-startup-delivery.md
new file mode 100644
index 0000000000..7220391f82
--- /dev/null
+++ b/docs/architecture/rfcs/ledger/loopx-overall-roadmap-v0/2026-10-30-s2-s12-windows-startup-delivery.md
@@ -0,0 +1,160 @@
+# S2/S12 startup diagnostic: Windows delivery window (2026-10-30)
+
+Measured on `main` at `d3d9ad209d`. This entry records what the
+[2026-10-04 startup diagnostic
+checkpoint](2026-10-04-s2-s12-startup-diagnostic.md) did not establish on
+Windows, and corrects the premise that produced an earlier wrong conclusion
+about the startup budgets. It changes no runtime behavior.
+
+## What the earlier checkpoint established, and what it left open
+
+That checkpoint records that when the managed Effect runtime cannot publish its
+startup locator, the failure travels through the typed startup envelope and
+reuses the shared filesystem/lock codes, and that a real fixture holding a
+live mutation lock "fails exactly once with the exact code". Its evidence
+was a real fixture; it did not qualify Windows, and it explicitly did not
+claim that Windows CI passes.
+
+## Measured
+
+The Windows lane fails intermittently on operations that start the Effect
+runtime and wait for it to become ready. The failing set differs per run:
+`test_effect_runtime_integration`'s `live_lock` case,
+`test_shadow_observation_retirement`, `test_chat_manager_inspection`.
+
+Across six Windows lane runs whose job logs are still retrievable, the
+failing file set was `test_self_update_runtime_activation` alone (three runs),
+that file together with `test_shadow_observation_retirement` (one run), that
+pair together with `test_effect_runtime_integration` (one run), and
+`test_chat_manager_inspection` alone (one run).
+
+The `live_lock` case failed on Windows at `1e3401591` with
+`assert 'runtime_startup_timeout' == 'mutation_lock_timeout'`, and passed in
+the later Windows runs of this entry's own measurement. It carries no platform
+skip, so it does run on the Windows lane.
+
+The startup cost was measured on the Windows runner at `2f544f9692` and on
+macOS at `d3d9ad209d`:
+
+| Host | Ready startup | Live-lock total | Diagnostic delivered |
+| --- | --- | --- | --- |
+| Windows (`nt`) | 0.406 s | 5.813 s | `mutation_lock_timeout` |
+| macOS (`posix`) | 0.170 s | 5.140 s | `mutation_lock_timeout` |
+
+The two revisions differ in `effect_runtime.py`, but only in the runtime-source
+metadata read, `_control_plane_root` and `_request_with_info`. The startup
+path itself — `_start_runtime`, `STARTUP_READY_TIMEOUT_SECONDS`,
+`_read_startup_stderr` and the `process.poll()` check — is unchanged between
+them, so both hosts are measured on the same startup code.
+
+The live-lock total is the ready startup plus the TypeScript mutation-lock
+wait, and both budgets have ample headroom: `STARTUP_READY_TIMEOUT_SECONDS`
+is 15.0 and `MUTATION_LOCK_TIMEOUT_MS` is 5,000.
+
+**This corrects an earlier conclusion.** The hypothesis that "node cold start
+plus the 5 s lock wait exceeds the 15 s ready budget on Windows" is false. The
+cold start is 0.406 s, and the live-lock case returned the exact
+`mutation_lock_timeout` code on that same run. Neither budget should be
+widened to explain this failure.
+
+## Where the exact diagnostic is lost
+
+TypeScript computes the typed code and publishes it:
+`effect_runtime_server.ts` `failStartup` writes the startup envelope to
+stderr with `writeSync` and then calls `process.exit`. Python reads that
+envelope only when a poll *inside* the ready loop shows the child has already
+exited (`effect_runtime.py`); after the loop it polls once more, but that
+poll only decides whether to terminate the child, and the function then
+reports the generic `runtime_startup_timeout` without reading the capture
+again.
+
+So the exact code is produced but not delivered whenever the parent does not
+observe the child's exit inside the ready loop.
+
+The lock in this path is **not** a stale lock. The fixture writes the
+current process's own PID as the lock owner, so the owner is alive and
+`reclaimStaleMutationLock` correctly refuses to reclaim it; waiting out the
+5 s mutation-lock budget is the intended behaviour, on every platform. The
+owner-liveness check in `effect_runtime_io.ts` records only a PID, so a
+reused PID or an `EPERM` probe result would also read as alive, but that is
+not what this fixture exercises and this entry does not claim it as the
+cause here.
+
+The measured live-lock total is 5.813 s on Windows and 5.140 s on macOS
+against a 15 s ready budget, and the exact code was delivered on both. So
+the budget is not the constraint in the measured runs.
+
+The intermittent Windows failure at `1e3401591` is a different situation. Its
+job log is retrievable and shows
+`assert 'runtime_startup_timeout' == 'mutation_lock_timeout'`. That code is
+produced whenever the parent does not observe the child's exit inside the
+ready loop, and the historical run does not distinguish two situations: the
+child had not published yet when the budget expired, or the child had
+already published and exited and the parent had not observed that exit
+before it left the loop. The same job's pytest step ran 439 cases in 5 min
+31 s, and the chunk holding the failure took two minutes against six seconds
+for the chunk before it, so the runner was heavily loaded. That load is a
+plausible contributor to a late publication, not a demonstrated cause, and
+this entry does not settle which situation occurred.
+
+Reading the capture at the moment the budget expires recovers the envelope
+in the second situation only. Tracing one child process shows the ordering
+directly: at one second, three seconds and five seconds the child is still
+running and stderr holds zero bytes, and the envelope appears only once the
+child exits, at about 5.8 s, together with exit code 2. The TypeScript side
+reaches `failStartup` only after the mutation-lock budget is spent, so a child
+that has not exited has published nothing; a child that has exited leaves the
+envelope in the capture, and the baseline leaves it unread.
+
+It can be recovered by letting the child finish first, or by reading the
+capture again once the child has exited. Waiting for the exit instead of
+terminating at the budget yields the exact code in every measured case, and
+the extra wait is bounded rather than open-ended: the child's exit lands at
+5.13 s to 5.17 s regardless of whether the ready budget was 0.5 s, 1 s or 2 s,
+because what the child waits for is the TypeScript mutation-lock budget. A
+repair would therefore give the child a bounded grace window after the ready
+budget, read the envelope if it appears, and report the generic timeout only
+when the child does not finish inside that window. Such a repair covers both
+situations above, so it does not depend on knowing which one the recorded
+failure was. That repair is not made here.
+
+## What this entry does not establish
+
+No budget change is proposed or qualified; the measurements show both budgets
+are sufficient. No owner-liveness change is proposed here: the lock in this
+path is held by a live owner, so the owner-liveness check behaves correctly
+for it, and changing how an owner identity is recorded would be a runtime
+change with its own acceptance that this measurement does not justify.
+
+The intermittent Windows failure is now attributed to a parent that did not
+observe the child's exit inside the ready loop; the historical record does
+not distinguish a child that had not published yet from a child that had
+published and exited unobserved, and the heavily loaded runner is a
+plausible contributor rather than a demonstrated cause. A bounded grace
+window after that budget is the repair this entry identifies, and it covers
+both situations, so it does not depend on knowing which one occurred. The repair itself is
+not made here: giving the child a grace window changes how long a failed
+startup blocks its caller, and that needs its own acceptance and a caller
+that tolerates the longer wait.
+
+Two acceptance points follow from it. The caller now waits longer, so the
+grace window has to be bounded by something the repository already owns
+rather than by a new constant. And the user-visible guidance changes with
+the code: `loopx/cli_commands/start_goal.py` gives `runtime_startup_timeout`
+its own remediation text, while `mutation_lock_timeout` falls through to the
+shared default, so a caller that receives the exact code would be told to
+repair or reinstall LoopX where it was previously told that startup timed
+out. That difference is the intended consequence of recovering the exact
+code, and it has to be read as such rather than as a regression.
+
+The delivered typed code is a diagnostic improvement, not a correctness
+repair: the runtime still fails to publish its locator, and the caller still
+fails to start. Windows CI is not qualified as passing by this measurement,
+and the intermittent failure set is preserved rather than reclassified.
+
+The measurement itself was a temporary diagnostic, removed after the numbers
+were read; it is not a qualification artifact in this repository. The commit
+that produced the numbers is reachable as `codex/measurement-archive` on the
+fork, so the two figures can be re-derived rather than taken on trust: it
+raises a deliberate `AssertionError` carrying the measurement, because the
+Windows lane runs pytest without `-s` and would otherwise capture the output.
diff --git a/docs/architecture/rfcs/ledger/loopx-overall-roadmap-v0/2026-10-30-s2-s12-windows-startup-delivery.zh-CN.md b/docs/architecture/rfcs/ledger/loopx-overall-roadmap-v0/2026-10-30-s2-s12-windows-startup-delivery.zh-CN.md
new file mode 100644
index 0000000000..cc11731c5d
--- /dev/null
+++ b/docs/architecture/rfcs/ledger/loopx-overall-roadmap-v0/2026-10-30-s2-s12-windows-startup-delivery.zh-CN.md
@@ -0,0 +1,118 @@
+# S2/S12 启动诊断：Windows 送达窗口（2026-10-30）
+
+在 `main` 的 `d3d9ad209d` 上测量。本条目记录
+[2026-10-04 启动诊断检查点](2026-10-04-s2-s12-startup-diagnostic.zh-CN.md)
+在 Windows 上没有确立的部分，并纠正由此产生的一个关于启动预算的错误结论。
+本条目不改变任何运行时行为。
+
+## 旧检查点确立了什么，未确立什么
+
+那个检查点记录：当受管 Effect runtime 无法发布其启动 locator 时，故障通过类型化启动
+信封传递并复用共享的文件系统/锁码；并记录一个持有活 mutation lock 的真实 fixture
+"恰好失败一次并给出精确的码"。它的证据来自真实 fixture；它没有给 Windows 资格，
+并且明确没有声称 Windows CI 通过。
+
+## 实测
+
+Windows lane 在"启动 Effect runtime 并等待其就绪"这类操作上间歇失败。每次 run 的失败
+集合都不同：`test_effect_runtime_integration` 的 `live_lock` 用例、
+`test_shadow_observation_retirement`、`test_chat_manager_inspection`。
+
+在六次 job 日志仍可获取的 Windows lane run 中，失败文件集合分别是：只有
+`test_self_update_runtime_activation`（三次）；该文件加上
+`test_shadow_observation_retirement`（一次）；该两者再加上
+`test_effect_runtime_integration`（一次）；只有 `test_chat_manager_inspection`
+（一次）。
+
+`live_lock` 用例在 Windows 的 `1e3401591` 上以
+`assert 'runtime_startup_timeout' == 'mutation_lock_timeout'` 失败过，而在本条目
+自身测量的后来几次 Windows run 中通过。它没有平台 skip，因此确实会在 Windows lane
+上运行。
+
+启动成本分别在 Windows runner 的 `2f544f9692` 与 macOS 的 `d3d9ad209d` 上测量：
+
+| 宿主 | 就绪启动 | live-lock 总计 | 送达的诊断 |
+| --- | --- | --- | --- |
+| Windows (`nt`) | 0.406 秒 | 5.813 秒 | `mutation_lock_timeout` |
+| macOS (`posix`) | 0.170 秒 | 5.140 秒 | `mutation_lock_timeout` |
+
+两个 revision 的 `effect_runtime.py` 有差异，但只在运行时源码元数据读取、
+`_control_plane_root` 和 `_request_with_info`。启动路径本身——`_start_runtime`、
+`STARTUP_READY_TIMEOUT_SECONDS`、`_read_startup_stderr` 与 `process.poll()` 检查——
+在两者之间未变，因此两个宿主测的是同一份启动代码。
+
+live-lock 总计等于就绪启动加上 TypeScript 的 mutation lock 等待。两个预算都有充裕余量：
+`STARTUP_READY_TIMEOUT_SECONDS` 为 15.0，`MUTATION_LOCK_TIMEOUT_MS` 为 5,000。
+
+**本条目纠正一个此前的结论。** "Windows 上 node 冷启动加上 5 秒锁等待会超过 15 秒就绪
+预算"这一假设是错的。冷启动为 0.406 秒，且同一次运行中 live-lock 用例返回了精确的
+`mutation_lock_timeout`。两个预算都不应为了解释这个失败而放宽。
+
+## 精确诊断在哪里丢失
+
+TypeScript 算出了类型化码并发布它：`effect_runtime_server.ts` 的 `failStartup` 用
+`writeSync` 把启动信封写到 stderr，然后调用 `process.exit`。Python 只在就绪循环
+*内部*的轮询显示子进程已经退出时才读那份信封（`effect_runtime.py`）；循环结束后它
+还会再轮询一次，但那次轮询只用来决定是否终止子进程，随后就报告通用的
+`runtime_startup_timeout`，不会再去读捕获文件。
+
+因此只要宿主没有在就绪循环里观察到子进程的退出，精确码已经产生却没有送达。
+
+这条路径上的锁**不是**陈旧锁。fixture 把当前进程自己的 PID 写成锁属主，因此属主是
+活着的，`reclaimStaleMutationLock` 正确地拒绝回收它；等满 5 秒 mutation lock 预算
+是设计行为，在所有平台上都一样。`effect_runtime_io.ts` 的属主存活判定只记录 PID，
+因此被复用的 PID 或 `EPERM` 的探测结果也会被读成存活，但 fixture 并未触及这一点，
+本条目也不把它当作此处的原因。
+
+实测的 live-lock 总计在 Windows 上是 5.813 秒、macOS 上是 5.140 秒，而就绪预算是
+15 秒，两个宿主都送达了精确码。因此在实测的那几次运行里，预算不是约束。
+
+`1e3401591` 上那次间歇的 Windows 失败是另一种情形。它的 job 日志仍可获取，显示
+`assert 'runtime_startup_timeout' == 'mutation_lock_timeout'`。这个码在宿主没有于就绪
+循环里观察到子进程退出时产生，而那次历史运行无法区分两种情形：预算耗尽时子进程
+尚未发布，或者子进程其实已经发布并退出、而宿主在离开循环前没有观察到那次退出。
+同一个 job 的 pytest 步骤用 5 分 31 秒跑完 439 个用例，含失败的那一段用两分钟，
+而前一段只用六秒，说明 runner 负载很重。这种负载是"发布迟到"的一个合理成因，但不是
+被证明的根因，本条目不确定那次究竟属于哪一种。
+
+在预算耗尽的那一刻补读捕获文件，只能在第二种情形里拿回信封。跟踪一个子进程直接给出
+了时序：在 1 秒、3 秒和 5 秒，子进程仍在运行且 stderr 有 0 字节；信封只在子进程退出
+时才出现——约 5.8 秒，与退出码 2 同时。TypeScript 侧是在 mutation lock 预算耗尽之后
+才走到 `failStartup`，因此还没退出的子进程没有发布任何东西；而已经退出的子进程会把
+信封留在捕获文件里，基线却不去读它。
+
+但先让子进程结束、或在子进程退出后再读一次捕获文件，就能拿回。改为等待退出而不是
+在预算处终止，在每个实测情形里都拿到了精确码，而且这段额外等待是有上界而非无界的：
+无论就绪预算是 0.5 秒、1 秒还是 2 秒，子进程都在 5.13 到 5.17 秒退出，因为它等的是
+TypeScript 的 mutation lock 预算。因此修法是：在就绪预算之后给子进程一个有界的收尾
+窗口，读到信封就用它，只有子进程在这个窗口内仍未结束时才报告通用超时。这样的修法
+同时覆盖上面两种情形，因此它的收益不依赖于判定那次记录在案的失败究竟属于哪一种。
+这个修法不在本条目里实施。
+
+## 本条目没有确立什么
+
+本条目不提出也不认证任何预算变更——实测表明两个预算都足够。这里也不提出属主存活性
+变更：这条路径上的锁由一个存活属主持有，属主存活判定对它行为正确，而改变属主身份的
+记录方式是一次带自有验收的运行时变更，本次测量并不足以支撑它。
+
+间歇性 Windows 失败现在归因于：宿主没有在就绪循环里观察到子进程的退出；历史记录
+无法区分"子进程尚未发布"与"子进程已经发布并退出但没被观察到"，繁忙的 runner 只是
+一个合理成因而不是被证明的根因。本条目指出的修法是那个预算之后的有界收尾窗口，
+它同时覆盖这两种情形，因此不依赖于判定那次究竟属于哪一种。这个修法本身不在此实施：给子进程一个收尾
+窗口会改变一次失败启动阻塞调用方多久，这需要它自己的验收，以及能容忍更长等待的调用方。
+
+由此产生两个验收点。其一，调用方等待的时间变长了，因此收尾窗口必须由仓库已经拥有的
+某个量来界定，而不是引入一个新常量。其二，用户可见的指引会随诊断码一起改变：
+`loopx/cli_commands/start_goal.py` 为 `runtime_startup_timeout` 准备了专门的修复提示，
+而 `mutation_lock_timeout` 落到共享兜底提示上；因此收到精确码的调用方，之前被告知
+"启动超时"，之后会被告知"修复或重装 LoopX"。这个差异是取回精确码的预期后果，
+应当被这样理解，而不是当作回归。
+
+送达的类型化码是诊断信息上的改进，不是正确性修复：runtime 仍然无法发布其 locator，
+调用方仍然启动失败。本测量不构成 Windows CI 通过的资格，间歇失败集合被原样保留，
+没有被重新归类。
+
+测量本身是一次性诊断，读到数字后已移除；它不是留在本仓库里的资格产物。产生这两个
+数字的提交在 fork 的 `codex/measurement-archive` 分支上仍然可达，因此这两个数字可以被
+重新推导，而不必只能被采信：它用一个刻意抛出的 `AssertionError` 携带测量值，因为
+Windows lane 跑 pytest 时没有 `-s`，否则输出会被吞掉。