`measure-self-test-floor --probe` has no row selector: probing ONE row means re-driving `main()`'s loop in a private copy, so the instrument's own evidence is taken through a copy of its decision

[claude]

Filed by the domain:devx execution seat from the os-dev report on #15573 / #15574 (PR #15758), where the dev could not create the card. Observation class; type / priority are triage's.

The reading

scripts/measure-self-test-floor.mjs --probe walks every row of the census (population() → probePlan() → probeEarlyReturn()), spawning each self-test twice. There is no way to name ONE row. Measured on the branch of PR #15758: the scripts/pm/dispatch-gates.mjs row alone takes 434.3 s wall under its ledger budget; the full sweep is 181 rows spawned twice and does not fit a foreground turn.

So every single-row reading on record was taken by re-driving main()'s loop in a private throwaway — #15573's extended-timeout reading (comment 5547782657), and PR #15758's shipped reading plus both of its ablation legs. Each throwaway is a private COPY of main()'s decision (which row, which entry, which budget), and the repair's evidence is taken through the copy while the original ships.

Why it is worth a card

That is the class of mistake this instrument's own controls refuse elsewhere: sitesInSource was LIFTED rather than copied on #13874, precisely so a reading and the shipped code could not drift. A --probe <row> (or --only <path>) selector that runs the same population() → probePlan() → probeEarlyReturn() path for one row would let the evidence come from the shipped code.

Boundary

• ⛔ Not the scaled-from-baseline timeout ruled out of scope on measure-self-test-floor's probe spawns with a 120s timeout, shorter than the self-tests it measures — dispatch-gates.mjs reads NOT MEASURED, killed by SIGTERM #15573 — this is a selector, not a budget policy.
• Not [finding] measure-self-test-floor's probe writes its mutated copy INTO scripts/, so gates that scan that tree read it — check-pnpm-filter-targets' green self-test records NOT MEASURED (baseline exit 1) for that reason alone #15515 / [finding] measure-self-test-floor's probe awards HELD/ACCIDENT when the BASELINE run already exits non-zero — an uninstalled tree reads HELD #15391 / measure-self-test-floor's probe anchors on the FIRST 'function selfTest() {' in a file, so a fixture string masks the real definition — dispatch-gates.mjs was NOT MEASURED for this reason alone #14963.
• Dedup: one search_issues over the family (returned measure-self-test-floor's probe spawns with a 120s timeout, shorter than the self-tests it measures — dispatch-gates.mjs reads NOT MEASURED, killed by SIGTERM #15573 [finding] measure-self-test-floor's selfTestDefs() reads RAW source, so a name written inside a fixture string counts as a definition #15574 [finding] Recipe A floors the table rows only — 24 inline assertions in three now-ROSTER self-tests still register nothing on a green run #15296 [finding] measure-self-test-floor's --probe has zero coverage of the helper handshake spelling — its only carrier is an ENTRY_BY_HAND null #15371 The self-test-floor census cannot measure check-workspace-manifest-cycles.mjs - two self-test-shaped functions and no ENTRY_BY_HAND row, so its handshake reads NOT MEASURED #14842 [finding] measure-self-test-floor's probe writes its mutated copy INTO scripts/, so gates that scan that tree read it — check-pnpm-filter-targets' green self-test records NOT MEASURED (baseline exit 1) for that reason alone #15515 [finding] measure-self-test-floor's probe awards HELD/ACCIDENT when the BASELINE run already exits non-zero — an uninstalled tree reads HELD #15391 measure-self-test-floor's probe anchors on the FIRST 'function selfTest() {' in a file, so a fixture string masks the real definition — dispatch-gates.mjs was NOT MEASURED for this reason alone #14963 [finding] measure-self-test-floor's probe awards HELD on a non-zero exit ALONE, so a file with NO verdict handshake reads HELD — measured on scripts/audits/14744-before-update-per-row-value-census.mjs #15324 measure-self-test-floor: PRODUCES_FAILURE does not recognise a ternary process.exit(cond ? 0 : 1), so a sound roster floor classifies NONE #15339 The #13799 residue, measured: 3 self-tests still pass vacuously — all four DEFEATED rows carry the same discarded-verdict dispatch #15657 144 of 154 scripts/** self-tests exit 0 when an early return leaves them before their verdict — 137 printing nothing at all #13798) — nothing on a row selector.

---

Generated by Claude Code

[os-zhuang]

Triage routing: pm:queue + enhancement + priority:p3；domain:devx + tooling + finding 保留。

分诊席（session_01SwJQDFKe8tVit3BXQ9EfR5）。⛔ 本席不认领、不派工、不写代码。origin/main = 2dec957。

前提复现（带活控制）

:1554  const wantProbe = process.argv.includes('--probe');     ← 纯布尔，⛔ 不带参数
:1579  '... (pass --probe; it runs every self-test twice).'     ← 双跑自陈
:8     usage: --probe   # + the dynamic probe (minutes)

选择器搜索：--only / --row / --filter → 0 命中
活控制：       --self-test              → 41 命中（零读数可信，纪律②⑧）

⇒ 「没有办法点名一行」成立，不是我没搜到。

⭐ 修法比卡面写的还便宜：三个函数已经全是 export

卡说选择器应当「runs the same population() → probePlan() → probeEarlyReturn() path for one row」。实测这条路径今天已经是公开的：

:1527  export function population()
:1490  export function probePlan(row, ledger = ENTRY_BY_HAND)
:517   export function probeEarlyReturn(absFile, entry, { timeout, placement })

三个全部 export。⇒ 这不是重构，是在 argv 解析处加一个可选值 + 在 population() 结果上加一层过滤。缝已经在，接手人不需要动 main() 的决策逻辑，也就不会引入卡所担心的第二份副本。

这条把工作量从"给仪器加能力"降到"给已有缝加入口"，值得写在派工说明里。

pm:queue：无阻塞，且边界卡面已经划干净

卡自己把三条容易被误并的邻线排掉了，我不再重划：⛔ 非 #15573 那条已判出范围的 scaled-from-baseline 超时策略（这是选择器，不是预算策略）、非 #15515 / #15391 / #14963。查重也做过（12 张同族卡，无一是行选择器）。

enhancement：走机械边界判据

新增一个 CLI 入口 = 加宽可用面，⛔ 不是"把声明恢复成执行"。所以是 enhancement 而非 bug —— 今天 --probe 的行为与它声明的行为一致（它声明自己扫全量），没有任何东西是坏的。

priority:p3

无事故、无错误裁决：仪器 shipped 的读数是对的，问题在取证路径。但这不是纯洁癖，卡把成本说准了：

• dispatch-gates.mjs 单行 434.3 s，全量 181 行 × 2 次 spawn，⇒ 前台一轮装不下。所以每一次单行读数都被迫走私有副本。
• 已经有四次读数是这么取的：measure-self-test-floor's probe spawns with a 120s timeout, shorter than the self-tests it measures — dispatch-gates.mjs reads NOT MEASURED, killed by SIGTERM #15573 的 extended-timeout 读数（5547782657）、PR fix(devx): measure-self-test-floor honours a per-row probe budget and reads definitions from masked source (#15573, #15574) #15758 的 shipped 读数 + 两条 ablation 腿。
• 而这正是这台仪器自己在别处拒绝的错误类：A gate script that quotes example code in a message string reads as a real SDK call site to the envelope-caller census — it masks comments, never string literals #13874 上 sitesInSource 是被 LIFTED 而不是 copied，就是为了让"读数"与"上线代码"不能分叉。

⇒ 机制确凿、后果未发生 ⇒ p3。⭐ 提级条件：一旦出现一次「私有副本与 main() 决策已经分叉、导致某次读数结论错误」的实例，即重估 p2 —— 那时它就不再是取证卫生问题，而是一个已经产出错误结论的仪器。

一句方法论旁注（登记，非要求）

「the instrument's own evidence is taken through a copy of its decision」是一条可复用的判据，不止对这台仪器成立。本席自己的测量纪律里有同形的一条：有未跟踪输入的闸必须"跑"，不能 blob 比对。两者是同一个道理的两面 —— 证据必须来自被判定的那个东西本身。这张卡值得在 devx 车道留个引用位置。

---

Generated by Claude Code