finding(tooling): the qa/http-conformance TEST_DEBT re-measure already needs ~4.4 GB of tsc heap on main — it passes the 4096 MB ceiling only through GC pressure, so any spec declaration growth tips Type Check · debt ledger to OOM

Found while driving PR #14513 (#14439) to green. Filed for the maintainer because the ceiling is a pinned policy (CI_TSC_HEAP_CEILING_MB = 4096, #12856) and changing what the program checks changes what the ledger measures.

Measured

scripts/check-type-check-coverage.mjs --re-measure builds one tsc --noEmit program per ledger entry under --max-old-space-size=4096. For the packages/qa/http-conformance TEST_DEBT entry (8 source files, 5 of them tests, but importing @objectstack/core, @objectstack/objectql, @objectstack/runtime, @objectstack/plugin-hono-server, @objectstack/driver-sqlite-wasm, @objectstack/spec), the same program shape run with an 8 GB cap and --extendedDiagnostics:

	`main` @ `53d36892` (base worktree, closure built)
Files	906
Lines of Definitions	691,580
Types	1,801,110
Instantiations	7,342,314
Memory used (tsc's own count)	4,473,321 K
Check time	68.9 s

Under the 4096 MB ceiling the program still completes on main (merge-queue Lint & Type Check runs are green through 53d36892 and later), but only because V8 collects harder near the limit: a local run on a main-based worktree peaked at 3,961 MB RSS and spent minutes in GC before finishing. The headroom is therefore a few percent, not a margin.

PR #14513 showed what that means in practice: a spec change that added 42,622 lines of definitions to this program (the whole ObjectStackDefinitionSchema declaration, pulled in through a new chunk edge — fixed in the PR) moved the count to 4,875,249 K and the lane went red twice. The next spec PR that grows the declaration graph seen by @objectstack/spec/system will hit the same wall without having done anything wrong.

Options

	Option	Cost	Note
A	Raise `CI_TSC_HEAP_CEILING_MB` to 6144 for the re-measure	Policy change; the file's own comment says 4096 is pinned to the CI runner's measured old-space and "if 4096 is wrong, it is wrong DOWNWARD" — `ubuntu-latest` has 7 GB, so 6144 needs a measurement of what else runs in that job	Restores headroom without changing what is measured
B	Shrink the program: `skipLibCheck: true` in the re-measure project only	Changes the measurement semantics of every TEST_DEBT number (declaration files stop being checked); the frozen numbers would need a wholesale re-measure	Large heap win (declaration files dominate: 691k lines of definitions vs 17k of TypeScript)
C	Shrink the program's inputs: have `http-conformance` type its tests against narrower entries (it imports whole packages for a handful of names)	Per-package edit; recurs for the next heavy entry	Local, no policy change
D	Leave as is; treat each OOM as the offending PR's to fix	Every spec PR pays a debugging cycle for a ceiling it did not set	What happened on #14513

Not decided here. The four-axis reading favours A with a measured number (long-term consistency: the ceiling exists to make numbers portable, and a ceiling with 3% headroom makes them fragile instead; scope: one constant, one measurement of the runner). B is the only option that changes ledger semantics and should not ride a spec PR.

os-dev-report

{
  "issue": 14569,
  "status": "needs_decision",
  "branch": "claude/issue-14569-tsc-heap-ceiling-6144",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14805",
  "premise_still_valid": true,
  "summary": "The ruling's first half is done and landed: the re-measure heap ceiling was measured first-hand on the `Type Check · debt ledger` runner (temporary probe commit emitting `::notice` annotations, run 33708954003 / job 100504131338, read back over the check-run annotations REST endpoint, then reverted in full), and the reading is now written into the comment beside CI_TSC_HEAP_CEILING_MB. The reading CONFIRMS the finding's premise (cap 4096 gives 4,077,718K used and a 26.84s check; cap 6144 gives 4,420,706K used and a 21.90s check: 343 MB more heap keeps 343 MB more live and finishes ~5s sooner, i.e. under 4096 the program pays GC pressure to fit) and corrects one fact in the card: `ubuntu-latest` here is 16,373,452 kB (~15.6 GiB), not 7 GB, with only ~918 MB held by the 153 other processes and nothing running beside the re-measure (the job's steps are sequential). The ruling's second half -- pin 6144 -- is NOT taken, because the measurement shows it cannot be delivered by the ruled file surface: the runner's own V8 default old space is 4096 MB (`heap_size_limit` 4144 MB, read directly), `remeasureHeapCeiling` minimises over the pin and the running process's own limit, so a 6144 pin still chooses 4144 -- and the `stale` arm then refuses outright, exiting 1 before the first tsc. Reproduced locally against a 4144 MB process. Shipping the bare raise would turn the debt-ledger lane red on every PR and on main, so the raise is escalated with three options instead of guessed. What lands: one file, 63 changed lines, all of them comments (mechanically counted, 0 non-comment).",
  "tests": "At 24ee34f8d7 (final commit, after `git merge origin/main`). `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands` derived 19 families (17 by path + 2 by change KIND, gate-script edit). 17 green: check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:ratchet-remedy-authority, check:type-check-coverage, check:watch-hint-literal, check-ci-filter-parity, check-cross-package-test-inputs, check-self-test-wired, check-shard-attestation, check-whole-set-label-write, pm/bare-root-worklist --self-test (exit codes captured with redirect-then-capture, never through a pipe). `node scripts/check-test-completeness.mjs` exits 3 = its own NOT MEASURED code (no local vitest summary), not a finding. NARROWING, DECLARED: `pnpm check:type-check-debt` was not run locally -- it needs a full closure build plus ~4 min of 34 sequential tsc programs on a shared container, and the diff is comment-only (63 of 63 changed lines are `//`). Substitutes: `--self-test` green (48 semantic + 68 observation + 43 re-measure + 28 built-closure + 19 auto-lowering + 18 exit-code cases, including all seven remeasureHeapCeiling rows), `check:type-check-coverage` green, and the debt lane itself ran GREEN in CI on this branch at d415f4fa40 (job 100504131338) on a tree carrying MORE than this one. The gate script has no vitest pin suite of its own (git grep over *.test.ts finds only comment mentions); its test surface is `--self-test`, which check-self-test-wired verifies. Reverse verification of the raise, run from a COMMITTED state: file mutated with perl -pi, mutation proven on disk by counting both spellings (4096-pin 1 then 0, 6144-pin 0 then 1) before any run; then `GITHUB_ACTIONS=true NODE_OPTIONS=--max-old-space-size=4096 node scripts/check-type-check-coverage.mjs --re-measure` exited 1 with '::error:: ... pins a CI heap ceiling of 6144 MB, but THIS CI runner's own default is 4144 MB'; restored with `git checkout HEAD -- ABSOLUTE_PATH` and restoration proven by hash (02c3b4bca638f1e5efb487ff01229d90a9de250c equals the HEAD blob), `git diff HEAD` empty, `git status --porcelain` empty. Control-byte scan of the edited file: grep -naP over the control ranges, exit 1 (no matches), captured before any pipe. CI on 24ee34f8d7 at report time: guards green (Part-of Closing-Keyword, Duplicate Fix, Single-Claim Path, Governed Surface, Check Links, PR Automation); `Lint & Type Check` and `CI` in_progress.",
  "mcp_calls": "0 -- REST only, with $GITHUB_TOKEN and --cacert; no MCP GitHub calls this run",
  "open_questions": [
    {
      "question": "The ruled raise to 6144 cannot be delivered by the constant, its comment and the self-test rows alone: on the measured runner the gate process's own V8 default old space is 4096 MB (heap_size_limit 4144), remeasureHeapCeiling minimises over the pin and that limit, and the `stale` arm refuses any pin above it (exit 1 before the first tsc, reproduced). How should the raise be delivered?",
      "options": [
        "A1 -- give the gate PROCESS the memory: `NODE_OPTIONS: --max-old-space-size=6144` on the re-measure step of the typecheck-debt job in lint.yml, then pin 6144. One workflow line plus the constant. Every invariant survives: the pin still never exceeds what the process really has, and the existing `stale` arm keeps the two numbers honest, loudly, on the runner itself. Cost: the ceiling is stated in two places (which the stale arm already polices) and it is OUTSIDE the ruled file surface, which is why it was not taken here.",
        "A2 -- change the cap rule: derive the second candidate from the box's REAL memory (MemTotal/MemAvailable) instead of this process's V8 default. Architecturally the more correct fix: that candidate exists to stop the gate promising memory the BOX does not have, and today it approximates that with V8's default, which is wrong on any box bigger than a 4 GB default heap -- measured here at 15.6 GiB with 10.1 GiB still free at the tightest moment of the heaviest program. Cost: it inverts the 'never RAISE' doctrine the file's header defends with two measured incidents, needs new adversarial self-test rows, and reopens the exit-137 SIGKILL failure mode on small containers unless the headroom maths is right.",
        "A3 -- do not raise. The pin describes the runner exactly, as measured; take option C from the finding (narrow qa/http-conformance's imports) so the heaviest program shrinks instead. Cost: recurs for the next heavy ledger entry."
      ],
      "recommendation": "A1. Long-term soundness (the axis carrying at least half the weight): the ceiling's whole job is to be a DESCRIPTION of what the process running tsc really has, and A1 is the only option that raises the number while keeping it a description -- the pin stays at or below the process's limit and the existing `stale` refusal keeps the two in step, on the runner, loudly. A2 is the better shape eventually but it is a policy inversion of a doctrine this file defends with measured incidents, so it deserves its own decision rather than riding this one. Real business need: measured and real -- the premise is confirmed (GC pressure, not comfort, at 4096) and #14513 already paid two red lanes for it. Making AI-written code hard to get wrong: A1 keeps the loud refusal that catches the two numbers drifting apart, while A3 leaves the next spec PR to discover the wall itself. Startup scope discipline: A1 is one workflow line plus one constant, no new surface. Whichever is chosen, the bare raise (constant only) must NOT ship: it reds the debt-ledger lane on every PR and on main."
    }
  ],
  "out_of_scope_findings": []
}

Generated by Claude Code

决策卡 — 裁定 A 的「提到 6144」在裁定的文件面内交付不了:门禁自己会拒绝任何高于进程 V8 默认堆的 pin

PM(devx 执行席,session session_01WLJQhde67SeTccsmnBVarV)按 dev 报告(os-dev-report,PR #14805)立卡。裁定 5519394826 的前半已完成并落在 PR #14805(draft,不挂 auto-merge):在 Type Check · debt ledger 的真实 runner 上第一手测量(临时探针提交 → check-run annotations 读回 → 探针全量 revert),读数已写进 CI_TSC_HEAP_CEILING_MB 旁的注释(63 行全是注释,0 行非注释)。后半(pin 6144)没有执行,原因是测出了对向事实:

runner 实测:ubuntu-latest MemTotal 16,373,452 kB(≈15.6 GiB,不是卡上写的 7 GB),job 里其它 153 个进程合计 ≈918 MB,再测量步骤是串行的、旁边没有别的消费者。
前提确认:cap 4096 → tsc Memory used 4,077,718 K、check 26.84 s;cap 6144 → 4,420,706 K、21.90 s。多给 343 MB 就多留 343 MB 活对象、快 5 s —— 4096 下确实是靠 GC 压力硬挤过去的。
拦路的事实:门禁进程自己的 V8 默认 old space 是 4096 MB(heap_size_limit 4144 MB,直接读出);remeasureHeapCeiling 取 pin 与进程自身上限的最小值,pin 写 6144 仍然选 4144;而 stale 分支对任何高于进程上限的 pin 直接拒绝——exit 1,连第一个 tsc 都不跑(dev 在本地对 4144 MB 进程复现)。裸改常量会让 debt-ledger 车道在每个 PR 和 main 上全红。

按 pm-dispatch「裁决明令的动作实施中测出对向事实 ⇒ 照裁决字面执行、冲突立成 needs-user-decision 卡、该 PR 不挂 auto-merge 留异议窗口」处置。三条可选交付路径(dev 给出,PM 复核):

	路径	改动	代价
A1	给门禁进程内存:在 `lint.yml` 的 typecheck-debt job 再测量步骤加 `NODE_OPTIONS: --max-old-space-size=6144`,然后 pin 6144	一行 workflow + 一个常量	上限写在两处(现有 `stale` 分支恰好就是盯这两处一致的);超出裁定的文件面(`.github/**`,非 governed)
A2	改上限规则:第二候选改为从机器真实内存(MemTotal/MemAvailable)推导,而不是本进程的 V8 默认	门禁逻辑 + 新的对抗性 self-test 行	反转文件头用两次实测事故守着的「never RAISE」信条;小容器上重开 exit-137 SIGKILL 失败面,除非余量算对
A3	不提:pin 按实测精确描述 runner;改走卡上的 C(收窄 `qa/http-conformance` 的 import)让最重的程序变小	单包改动	下一个重的台账条目还会再来一次

① 项目长远合理性(权重 ≥50%,领起推荐) —— 这个常量的全部职责是描述跑 tsc 的进程真实拥有的堆。A1 是唯一「把数字提上去、同时让它仍然是描述」的路径:pin 不超过进程上限,现有 stale 拒绝分支继续在 runner 上大声守住两数一致。A2 形状上更对(那个候选存在的目的是防门禁承诺机器没有的内存,今天却用 V8 默认近似,在任何比 4 GB 默认堆大的机器上都不对),但它是对文件自己用实测事故守着的信条的反转,应当单独立决,不该搭这张卡。A3 让 pin 保持诚实但把墙留给下一个 spec PR 自己撞。
② 实际业务拉动 —— 真实且已量化:前提在真实 runner 上被确认(4096 下是 GC 压力不是舒适余量),feat(cli+spec): compile a project of N packages into one packages[] artifact, with the assembled package body declared (ADR-0130 D4 producer, #14242 B) #14513 已经为此付过两次红车道;spec 声明每周都在长,触发是「何时」不是「是否」。
③ 防 AI 犯错 —— A1 保留「两数漂移就大声拒绝」的机制(而且这次 dev 正是被它拦住的,证明它有效);裸改常量正是会被下一个 AI 作者顺手做出来的错——所以无论选哪条,裸提常量 ⛔ 不得合入。A2 的余量算法如果写错,失败形态是 exit-137,读起来像基础设施而不是发现。
④ 创业阶段不扩散 —— A1 一行 workflow + 一个常量,零新面;A2 新增门禁逻辑 + 测试行 + 一道要重新论证的信条;A3 是逐包返工。

推荐:A1。 ①③④同向,②不构成反对。同一 PR(#14805)继续承载:一行 NODE_OPTIONS、常量 6144、pin 常量的 self-test 行,注释里的读数已在。
回退:A3(若维护者不愿让 workflow 与常量两处写同一个数)。
置信缺口: 没有量 6144 老空间在 runner 上对「其它消费者」的实际挤占(实测旁边没有并行消费者,所以缺口是理论上的);A2 的 headroom 公式没有设计,列出只为说明它值得单独一张卡。

标签同笔:pm:dispatched → needs-user-decision(assignee 清空;dev 的 worktree 与分支 claude/issue-14569-tsc-heap-ceiling-6144 保留,PR #14805 保持 draft 直到裁定)。

Generated by Claude Code

Maintainer ruling recorded — A1: give the gate process the memory, then pin 6144 (2026-09-03)

Provenance. Ruled by the maintainer (pm@objectstack.ai) in the director seat's live session (session_01ShyhexkB2d1AeRZ85tgAAe), decision batch #22, 2026-09-03 after 15:35Z. Verbatim reply: 「同意」. The batch was presented with this card's recommendation as A1 (fallback A3; A2 deserves its own decision and is not filed), so 「同意」 adopts A1. Ledger: objectstack#12708, batch #22 ledger comment.

What was ruled. The 02:25Z ruling A (comment 5519394826) stands — raise on a recorded measurement — and its delivery is completed the only way the measurement allows:

the re-measure step of the typecheck-debt job in .github/workflows/lint.yml sets NODE_OPTIONS: --max-old-space-size=6144, so the process running tsc really has the old space the pin describes;
CI_TSC_HEAP_CEILING_MB moves 4096 → 6144, with the self-test rows that pin it; the runner measurement already written beside the constant in PR fix(tooling): raise the re-measure heap ceiling to 6144 and give the job the memory (A1) #14805 stays as the evidence;
the stale arm keeps refusing a pin above the process's own limit — that refusal is what caught the bare raise, and it stays.
⛔ The bare constant raise never ships alone (it reds the debt-ledger lane on every PR and on main). ⛔ A2 (derive the cap from box memory) is not taken here; it inverts the file's "never RAISE" doctrine and would need its own card. A3 not taken.

Execution. domain:devx lane; PR #14805 continues on its branch. .github/workflows/** is not a governed surface: review, then auto-merge. Clause-②: no. skip-changeset.

State. needs-user-decision → pm:queue in the same stroke; tooling, priority:p2, domain:devx kept; unassigned.

Generated by Claude Code

Claim: PM loop round R4 — continuation on the existing branch, not a fresh start
Session: session_012zGPuVVX3deAx9LdjK8jCk
Branch: claude/issue-14569-tsc-heap-ceiling-6144 (existing; PR #14805, head 24ee34f8d7, last touched 03:06Z — before the ruling)
Worktree: objectstack-issue-14569
Domain: domain:devx
File surface: scripts/check-type-check-coverage.mjs + .github/workflows/lint.yml (stop on breach; explain in the report)
Container & model: S/M, mode:subagent, model: opus (a bare constant raise reds the debt-ledger lane on every PR and on main; the two halves must ship together or not at all — judgment, not substitution)
Clause-②: no (per the ruling)
Serial constraints cleared: none — 0 devs live, 0 in-flight PRs of this seat. ⚠️ scripts/check-type-check-coverage.mjs is a contended file: #13798 batch 3 wants it and #14943 is queued behind it. Neither is dispatched, and neither will be while this is in flight.

Decision re-read to the last page. Ruled A1 at 15:53:57Z (5528367074, director seat, decision batch #22, maintainer 「同意」). The 02:25Z ruling A stands — raise on a recorded measurement — and A1 completes its delivery the only way the measurement allows. ⛔ A2 (derive the cap from box memory) is not taken: it inverts the file's own never-RAISE doctrine and needs its own card. ⛔ A3 not taken.

⚠️ PR #14805 already exists and carries only the measurement (1 file, +62/−1, all of it the comment block written in the earlier round). The ruling says it continues on its branch — so this is a continuation. ⛔ Do not open a second PR, ⛔ do not restart the branch.

⭐ The standing rule that broke #15085 applies squarely here: a derived gate family is not a complete account of what CI runs. This diff edits check-type-check-coverage.mjs, a gate script — so it owes that script's own suite whether or not the derivation names it.

Generated by Claude Code

Claim (dev seat) — continuation of PR #14805 on its existing branch, not a fresh start.

Session: session_012zGPuVVX3deAx9LdjK8jCk (os-dev, subagent lane, model opus)
Branch: claude/issue-14569-tsc-heap-ceiling-6144 (existing; PR fix(tooling): raise the re-measure heap ceiling to 6144 and give the job the memory (A1) #14805 stays draft)
Worktree: ../objectstack-issue-14569, created off the existing branch head 24ee34f8d7; origin/main 369da918d merged in as a real merge commit (no rebase, no force-push, no amend). Write-route probe: that merge commit pushed clean before any edit.
File surface: scripts/check-type-check-coverage.mjs + .github/workflows/lint.yml, exactly as ruling A1 names them. No third file.
Ruling followed: A1 (comment 5528367074, 15:53:57Z, director seat, decision batch chore(ci)(deps): bump actions/upload-artifact from 4 to 6 #22, maintainer 「同意」), which completes ruling A (5519394826) rather than replacing it. The two halves ship together: the NODE_OPTIONS on the re-measure step and the CI_TSC_HEAP_CEILING_MB raise. The stale arm is untouched.
Not taken: A2 (deriving the cap from box memory) and A3, per the ruling.

Report follows on this card as os-dev-report.

Generated by Claude Code

fix(tooling): pin the re-measure heap ceiling at 6144 and give the jo…

os-dev-report

{
  "issue": 14569,
  "status": "done",
  "branch": "claude/issue-14569-tsc-heap-ceiling-6144",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14805",
  "premise_still_valid": true,
  "summary": "Ruling A1 delivered in full on the existing branch and the existing PR #14805 (no second PR, no restart, no rebase, no force-push): origin/main 369da918d merged in as a real merge commit cc3a72ae6, then one implementation commit a68aed7ec. BOTH HALVES ARE IN THE DIFF. (1) .github/workflows/lint.yml, job typecheck-debt, step 'Re-measure the type-check DEBT / TEST_DEBT ledger' now carries env NODE_OPTIONS: --max-old-space-size=6144 -- the step is exactly where the ruling said it is, unmoved by today's main (PM assumption 2 CONFIRMED, and the parsed YAML is quoted in the report's tests field). (2) CI_TSC_HEAP_CEILING_MB 4096 -> 6144 in scripts/check-type-check-coverage.mjs, plus two new remeasureHeapCeiling self-test rows pinning both directions of the pairing. The stale arm is byte-for-byte untouched and was NOT weakened; it is what makes the pair inseparable. Of 115 changed lines in the gate script exactly 16 are non-comment (mechanically counted) and they are the constant plus the two rows; the rest is the comment block. THE RUNNER MEASUREMENT IS STILL VALID: it was taken first-hand inside the Type Check debt ledger job earlier today (run 33708954003 / job 100504131338, image ubuntu24 20260831.293.1, Node v22.23.2, 4 vCPU) and today's main merge touches neither the runner image, nor the Node version, nor that job's shape -- it is re-stated in place as the pin's evidence, not re-taken (PM assumption 1 CONFIRMED). PM assumption 3 CONFIRMED and it was already measured, not assumed: the heaviest ledger program was RUN on that runner under --max-old-space-size=6144 (4,420,706K used, 4,545,500 kB peak RSS, 21.90s, no OOM) against 4,077,718K / 26.84s under 4096, with the lowest MemAvailable at any moment of either run 10,562,192 kB -- about 1.7x the 6144 MB the pin now asks for, on a 16,373,452 kB box with ~918 MB held by the job's other 153 processes. One consequence I did NOT find pre-recorded anywhere and have written into both files rather than leave for someone to rediscover: with NODE_OPTIONS set explicitly on that step, heap_size_limit there reads 6192 whatever the runner's physical memory does, so on that one job the stale arm can no longer notice the runner SHRINKING -- it now only notices a pin above a DEFAULTED process. That is a real (small) loss of a tripwire, the ~1.7x margin above is what makes it acceptable, and the constant's comment names the remedy if the margin is ever in doubt (a fresh runner measurement and a smaller number in BOTH places, never a bigger one here). PR #14805 left as a draft with auto-merge NOT armed, per the dispatch.",
  "tests": "ALL of the following at a68aed7ec, the final commit and the pushed head; every exit code captured by redirect-then-capture, never read through a pipe.\n\nTHE PAIRING (the assertion the ruling is built on), run on the committed tree, no mutation needed because the pairing is a property of the PROCESS's limit:\n  A. GITHUB_ACTIONS=true NODE_OPTIONS=--max-old-space-size=4096 node scripts/check-type-check-coverage.mjs --re-measure  -> EXIT=1, '::error::scripts/check-type-check-coverage.mjs pins a CI heap ceiling of 6144 MB, but THIS CI runner's own default is 4144 MB -- the pin is now ABOVE the ceiling it claims to describe', refused BEFORE the first tsc. That is the bare-raise failure this card exists to avoid, reproduced.\n  B. GITHUB_ACTIONS=true NODE_OPTIONS=--max-old-space-size=6144 node scripts/check-type-check-coverage.mjs --re-measure  -> stale arm SILENT, prints 'heap: tsc runs under --max-old-space-size=6144 MB -- the CI-shaped ceiling pinned by scripts/check-type-check-coverage.mjs; this process's own limit is 6192 MB', then EXIT=3 = PREREQUISITE NOT MET on an unbuilt closure (this gate's own NOT-MEASURED code, not a pass and not a red) -- i.e. it got PAST the arm, which is the whole point. Both directions are now also pinned as self-test rows so the pairing cannot rot silently.\n\npnpm check:type-check-debt -- RUN, NOT NARROWED (last round's narrowing does not carry: this diff moves the number that family uses). Closure build first, through the shared verify lock: pnpm exec turbo run build --filter='./packages/*' --filter='./packages/*/*' --concurrency=2 -> 71 successful, 71 total, VERDICT command-exit 0, held the lock 440s. Then the gate, also through the lock: '✓ check:type-check-coverage --self-test — 48 semantic + 68 observation + 45 re-measure + 28 built-closure + 19 auto-lowering + 18 exit-code case(s) hold.' / 'heap: tsc runs under --max-old-space-size=6144 MB -- the CI-shaped ceiling pinned by ...; this process's own limit is 8240 MB.' / 'check-type-check-coverage --re-measure: OK — 17 ledger entr(ies) re-measured in 138.9s, 217 raw tsc error(s) total, none above its recorded number.' DEBT_EXIT=0, VERDICT command-exit 0, held the lock 140s. The re-measure case count moved 43 -> 45, which is the two new rows.\n\nDERIVED FAMILY: node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (no paths passed -- it derives its own change set from the merge base 369da918d) reported 'gate list derived from the tree ... at commit a68aed7ec' and 45 families = 38 by path + 2 by change KIND (adds/edits a GATE SCRIPT) + 6 declared whole-tree, 1 reached both ways. Harvested with --commands (45 lines) and ALL 45 run: 44 exit 0; 1 exit 3 = node scripts/check-test-completeness.mjs, its own declared NOT MEASURED code (no local vitest summary), read as neither pass nor red. Green includes check:type-check-coverage, check:type-check-debt, check:pm-dispatch-gates, pm/bare-root-worklist --self-test, check-self-test-wired(+--self-test), check-self-test-workflow-commands(+--self-test), check-step-collectors(+--self-test), check-aggregator-roster(+--self-test), check-required-contexts, check-ci-filter-parity, check-closing-keyword-parity(+--self-test), check-comment-mask-corpus, check-shard-attestation, check-position-name-fold-loaders(+--self-test), check-whole-set-label-write(+--self-test), check-cross-package-test-inputs, docs-audit/check-drift-comment, pm/ci-failure --self-test, check:nul-bytes, check:refd-timer-probe, check:watch-hint-literal, check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:declared-population-live, check:entry-guard, check:node-version, check:parse-guard, check:pnpm-acquisition, check:pnpm-filter-targets, check:ratchet-remedy-authority, check:required-contexts, check:shard-attestation, check:stall-guard-budget, check:stall-guard-headroom, check:workflow-status-functions.\n\nTHE GATE SCRIPT'S OWN SUITE, owed whether or not the derivation names it (the rule that broke #15085): scripts/check-type-check-coverage.mjs is a gate script. Its test surface is --self-test, run green twice (standalone and inside check:type-check-debt), and check-self-test-wired confirms it is wired. git grep for the script's name across *.test.ts / *.test.mjs / *.test.js / *.spec.ts finds 12 files and EVERY hit is a prose mention inside a comment -- I opened all 12 and none executes the script. There is no vitest pin suite to have missed. check:pm-dispatch-gates ran green explicitly, not only as part of the family.\n\nWORKFLOW HALF, verified structurally rather than by eye: python3 yaml.safe_load of .github/workflows/lint.yml parses (6 jobs) and the step reads exactly {'name': 'Re-measure the type-check DEBT / TEST_DEBT ledger', 'env': {'NODE_OPTIONS': '--max-old-space-size=6144'}, 'run': 'pnpm check:type-check-debt'}. A step-level env: block is an established idiom in this repo (ci.yml carries two).\n\nESLINT -- NARROWING, DECLARED, with all three readings, so it is a measurement and not a skip: (1) population read from ESLint's OWN config, not guessed -- isPathIgnored('.github/workflows/lint.yml') = true (the workflow half is not in ESLint's population at all), isPathIgnored on the .mjs = false; (2) file count from --format json = 1 file linted, 0 errors, 0 warnings, ESLINT_EXIT=0; (3) config invariance for untouched files -- calculateConfigForFile reports languageOptions.parserOptions.project = undefined, i.e. type-aware linting is OFF, so this diff cannot move the verdict of any file it does not touch. The repo-wide `pnpm lint` is CI's run.\n\nCONTROL BYTES: grep -naP over [\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f] across both edited files -> exit 1, zero-byte output, captured before any pipe.\n\nEDIT-LANDED-ON-DISK PROOFS (every edit was applied by an anchor-counting python script that exits non-zero on a miss, and each was then observed, never trusted to an exit code): 'CI_TSC_HEAP_CEILING_MB = 6144' grep -c = 1 and 'CI_TSC_HEAP_CEILING_MB = 4096' grep -c = 0; the new self-test row label grep -c = 1; 'max-old-space-size=6144' present in lint.yml at line 4837. git status --porcelain is empty at the pushed head.\n\nCI at report time on a68aed7ec: 31 check runs -- 12 success, 8 skipped, 11 in_progress, 0 failing. Reporting now per the draft-PR-time contract; I did not wait for convergence.",
  "mcp_calls": "2 -- issue_read get and issue_read get_comments on #14569. Everything else went through git, the public-repo page payload (card body + front timeline), and REST with $GITHUB_TOKEN + --cacert (probed first: GET /repos/.../pulls/14805 -> HTTP 200), including the claim comment, this report comment, the PR body PATCH and the check-run read.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — PR #14805, ruling A1 delivered in full. ⭐ And it found a cost of the ruled change and wrote it into the code

domain:devx seat, session session_012zGPuVVX3deAx9LdjK8jCk, R4, 2026-09-03T23:5xZ. Checked against the diff, ⛔ not the report.

Both halves are in the diff — verified, and the ruling turns on them shipping together

half	reading on the branch
workflow	`.github/workflows/lint.yml:4835–4837` — step "Re-measure the type-check DEBT / TEST_DEBT ledger" now carries `env: NODE_OPTIONS: --max-old-space-size=6144`
constant	`CI_TSC_HEAP_CEILING_MB = 6144` present 1×, the old `4096` present 0×
the `stale` arm	⭐ untouched — I diffed the whole gate script against the merge base and filtered to non-comment lines: the only changes are the constant and two self-test rows. No arm logic moved

⭐ The two new rows pin the pairing in BOTH directions, so the coupling cannot rot silently:

"the workflow's own NODE_OPTIONS ties the pin and is not read as a tighter caller cap" → stale: false
"the same runner WITHOUT the workflow NODE_OPTIONS — its 4144 MB default — is refused" → stale: true

The evidence I asked for, delivered as a live pairing rather than a green run

NODE_OPTIONS=4096 ⇒ EXIT 1, refused before the first tsc: "pins a CI heap ceiling of 6144 MB, but THIS CI runner's own default is 4144 MB — the pin is now ABOVE the ceiling it claims to describe." ⇒ the bare-raise failure this card exists to prevent, reproduced.
NODE_OPTIONS=6144 ⇒ the arm goes silent and execution gets past it, ending at EXIT 3 (PREREQUISITE NOT MET on an unbuilt closure — the gate's own NOT-MEASURED code, neither pass nor red).

⇒ The raised pin is honest only with the workflow change, demonstrated rather than argued.

Premises, all three confirmed with measurements rather than assurances

The runner measurement is still valid — taken first-hand today inside the actual Type Check debt-ledger job (run 33708954003, image ubuntu24 20260831.293.1, Node v22.23.2, 4 vCPU); today's main merge touches neither the image, the Node version, nor that job's shape. Re-stated in place as evidence, ⛔ not silently re-taken.
The step is where the ruling said it is, unmoved by today's main — and confirmed structurally, by yaml.safe_load parsing the file and reading the step back, not by eye.
6144 is actually achievable — the heaviest ledger program was run on that runner under 6144 (4,420,706K used, 4,545,500 kB peak RSS, no OOM) against 4,077,718K under 4096, with the lowest MemAvailable at any moment ≈ 1.7× the 6144 the pin asks for.

⭐ The best thing in this delivery is a cost it found and refused to hide

With NODE_OPTIONS set explicitly on that step, heap_size_limit there reads 6192 regardless of the runner's physical memory — so on that one job the stale arm can no longer notice the runner shrinking. It now only notices a pin above a defaulted process.

That is a real, if small, loss of a tripwire, and it is a consequence of the ruled change rather than of the implementation. The dev found it, judged the ~1.7× margin sufficient to accept it, and wrote it into both files with the remedy named (a fresh runner measurement and a smaller number in both places, ⛔ never a bigger one here) instead of leaving it to be rediscovered by whoever next reads a confusing green.

The always-runs-tail rule was honoured — its first test

This diff edits a gate script, so per the rule verified on #15085 it owes that script's own suite whether or not the derivation names it. Done: --self-test green twice (standalone and inside check:type-check-debt), check-self-test-wired confirms it is wired, check:pm-dispatch-gates run explicitly and green, and a git grep across *.test.*/*.spec.ts found 12 hits — all 12 opened, every one a prose mention in a comment, so there is no vitest pin suite that was missed. ⭐ check:type-check-debt was run, not narrowed, correctly: last round's narrowing does not carry, because this diff moves the very number that family consumes.

The ESLint narrowing is a measurement, not a skip: population read from ESLint's own config (isPathIgnored — the workflow file is not in its population at all), file count from --format json, and config invariance shown by parserOptions.project === undefined (type-aware linting off), so this diff cannot move any untouched file's verdict.

45 derived families run: 44 exit 0, 1 exit 3 declared NOT MEASURED.

Landing

Governed check: get_files → .github/workflows/lint.yml + scripts/check-type-check-coverage.mjs, zero governed paths — .github/workflows/** is not governed, exactly as the ruling states ⇒ normal path.
CI on a68aed7ec at 23:56Z: 18 ✓ / 8 skipped / 5 in progress / 0 red ⇒ ⛔ not armed yet. Flip + arm on full green, every check.
On merge this closes #14569 (Fixes) and frees scripts/check-type-check-coverage.mjs for #13798 batch 3 and #14943, both of which have been held behind it.

Generated by Claude Code

LANDED — PR #14805 MERGED 00:29:24Z. Ruling A1 is in `main`, both halves together

domain:devx seat, session session_012zGPuVVX3deAx9LdjK8jCk, R4. Probed on origin/main after a re-fetch (9221ec8da).

probe	reading
workflow half	✅ `.github/workflows/lint.yml:4835–4837` — step "Re-measure the type-check DEBT / TEST_DEBT ledger", `env: NODE_OPTIONS: --max-old-space-size=6144`
constant half	✅ `CI_TSC_HEAP_CEILING_MB = 6144` present 1×; the old `4096` 0×
the `stale` arm	✅ survived intact — still refuses a pin "ABOVE the ceiling it claims to describe"

⇒ The ruling's core condition holds on main: the bare constant raise never shipped alone. Fixes #14569 auto-closed the card; pm:dispatched stripped and assignee cleared, since auto-close strips neither.

⭐ Recorded because it is a cost of the ruled change, not of the implementation

With NODE_OPTIONS pinned on that step, heap_size_limit there now reads 6192 regardless of the runner's physical memory — so on that one job the stale arm can no longer notice the runner shrinking; it only catches a pin above a defaulted process. The dev found this, judged the measured ~1.7× headroom sufficient (lowest MemAvailable 10,562,192 kB against the 6144 MB asked), and wrote it into both files with the remedy named: a fresh runner measurement and a smaller number in both places — ⛔ never a bigger one here.

Unblocked by this landing

scripts/check-type-check-coverage.mjs is now free. #13798 batch 3 and #14943 were both held behind it and are dispatchable.

ACCEPT: 5533671487.

Generated by Claude Code

fix(tooling): raise the re-measure heap ceiling to 6144 and give the …

Navigation Menu

Repository navigation

finding(tooling): the `qa/http-conformance` TEST_DEBT re-measure already needs ~4.4 GB of tsc heap on `main` — it passes the 4096 MB ceiling only through GC pressure, so any spec declaration growth tips `Type Check · debt ledger` to OOM #14569

Measured

Options

决策卡 — 裁定 A 的「提到 6144」在裁定的文件面内交付不了:门禁自己会拒绝任何高于进程 V8 默认堆的 pin

Maintainer ruling recorded — A1: give the gate process the memory, then pin 6144 (2026-09-03)

ACCEPT — PR #14805, ruling A1 delivered in full. ⭐ And it found a cost of the ruled change and wrote it into the code

Both halves are in the diff — verified, and the ruling turns on them shipping together

The evidence I asked for, delivered as a live pairing rather than a green run

Premises, all three confirmed with measurements rather than assurances

⭐ The best thing in this delivery is a cost it found and refused to hide

The always-runs-tail rule was honoured — its first test

Landing

LANDED — PR #14805 MERGED 00:29:24Z. Ruling A1 is in `main`, both halves together

⭐ Recorded because it is a cost of the ruled change, not of the implementation

Unblocked by this landing

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(tooling): the qa/http-conformance TEST_DEBT re-measure already needs ~4.4 GB of tsc heap on main — it passes the 4096 MB ceiling only through GC pressure, so any spec declaration growth tips Type Check · debt ledger to OOM #14569

Description

Measured

Options

Activity

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

决策卡 — 裁定 A 的「提到 6144」在裁定的文件面内交付不了:门禁自己会拒绝任何高于进程 V8 默认堆的 pin

os-project-manager commented on Sep 3, 2026

Maintainer ruling recorded — A1: give the gate process the memory, then pin 6144 (2026-09-03)

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

ACCEPT — PR #14805, ruling A1 delivered in full. ⭐ And it found a cost of the ruled change and wrote it into the code

Both halves are in the diff — verified, and the ruling turns on them shipping together

The evidence I asked for, delivered as a live pairing rather than a green run

Premises, all three confirmed with measurements rather than assurances

⭐ The best thing in this delivery is a cost it found and refused to hide

The always-runs-tail rule was honoured — its first test

Landing

claude commented on Sep 4, 2026

LANDED — PR #14805 MERGED 00:29:24Z. Ruling A1 is in main, both halves together

⭐ Recorded because it is a cost of the ruled change, not of the implementation

Unblocked by this landing

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(tooling): the `qa/http-conformance` TEST_DEBT re-measure already needs ~4.4 GB of tsc heap on `main` — it passes the 4096 MB ceiling only through GC pressure, so any spec declaration growth tips `Type Check · debt ledger` to OOM #14569

LANDED — PR #14805 MERGED 00:29:24Z. Ruling A1 is in `main`, both halves together