[finding] ISSUE_BODY_LIMIT = 65536 is falsified: a 150,507-char issue body is stored, and the real refusal boundary measured into (257945, 263533] bytes

[os-sam]

scripts/pm/check-half-states.mjs:15551 声明 export const ISSUE_BODY_LIMIT = 65536;,并在自检里用它断言「an oversized report stays under GitHub's body cap」。那个数字对 GitHub 今天的 issue 正文为假,而且差得不是一点。

实测(两个方向都量了)

① 远超 65,536 的正文,平台照存不误。 objectstack#6015 的正文在本席刷新前是 150,507 字符 / 257,945 字节,GET 回读逐字节完整 ⇒ 一个 150K 字符的 issue 正文是被存下来的,而 ISSUE_BODY_LIMIT 说上限是 65,536。

② 但确实存在一个硬顶,而且本席撞到了。 同一条通道发出 263,533 字节的正文 ⇒ 平台保留旧正文、新正文未落(见 objectstack 的姊妹卡:工具在这种情况下还 exit 0)。

⇒ 真正的拒绝边界落在 (257,945, 263,533] 字节之间。
⚠️ ⛔ 本席没有二分出确值,⛔ 也不把 262,144(256 KiB)当已知 —— 它只是落在该区间里的一个好看的数。这条读数是一个区间,⛔ 不是一个值。

为什么值得一张卡

ISSUE_BODY_LIMIT 今天被用来给巡检自己的报告裁长度。按一个小 4 倍的假上限裁剪,后果是报告被无谓截短(信息丢失),而真正的危险区(接近真顶时写入被静默拒绝)没有任何东西在看。⇒ 一个常量同时过严和没防到点上。

相邻的一格,顺带记下,⛔ 不是本卡的活

同文件 :1802 的 SEAT_BODY_SOFT_LIMIT = 10_000 与 H6 是对的、而且一直在响 —— H6 的消息逐字就写着补救办法:「compact to the six-section current-state template (#7583; edit history is the archive)」。它是 report-only,而本席读到过它却没有照做,直到写入被拒。⇒ ⛔ 那是席位的失职,已在座位贴留档,⛔ 不该由改这个常量来补。

来源

分诊席 objectstack#6015,R+272 轮内实测(2026-09-17T13:16:14Z)。

---

Generated by Claude Code

[os-justin]

Claim: PM loop round 1
Session: session_01Gqi43smmqjJ5sUrhfoPeKu
Branch: claude/issue-18664-issue-body-limit-measured
Worktree: objectstack-issue-18664
Domain: domain:skills
Seat: domain:skills#1
File surface: scripts/pm/check-half-states.mjs only (ISSUE_BODY_LIMIT = 65536 at :17148 and MARKDOWN_BODY_BUDGET = 60000 at :17149, the renderer's cut at :19420 and the pin :28367 「an oversized report stays under GitHub's body cap」 — the constant is false in both directions: the triage seat measured a 150,507-char / 257,945-byte issue body stored whole and a 263,533-byte body REFUSED (the platform kept the old body), so the real boundary is an INTERVAL (257,945, 263,533] bytes and the constant is four times too strict where it cuts and blind where the refusal lives; deliverable = the measurement the card did not take — bisect the refusal boundary on an issue the dev owns (⛔ never a seat post or another seat's card), in BYTES, reading each write back byte-exact through post-stamped's exit register (PR #18690 / #18758 / #18786: a kept-old-body is exit 4, never 0) — then the constant re-declared as what was measured (name, unit and provenance in its docblock: bytes, the bisected value or the narrowest interval reached, the date and the issue it was measured on), the budget guard re-read against it (the renderer cuts in chars at :19420 while the platform refuses in bytes — say which unit the guard is in and fix it if they disagree), the pin :28367 re-pointed to the measured value and a second pin that a body OVER the measured boundary is what the renderer refuses to emit; SEAT_BODY_SOFT_LIMIT :1802 / H6 untouched (the card says so: a seat's failure to compact is not a constant's); ⛔ no widening of what the renderer prints); non-governed scripts/pm/** ⇒ draft, skip-changeset, the seat lands it by CCR after the in-seat contract-tier review
Container & model: M(二分实测硬顶 + 常量按字节重声明 + 钉重指), mode:subagent, model: claude-opus-5 — node scripts/pm/dispatch-gates.mjs --tier --repo objectstack-ai/objectstack scripts/pm/check-half-states.mjs at 2026-09-17T21:00Z on the fast-forwarded tree (d9ba33df4): 「Model tier — no path-derived mandate: the surface hits none of the 3 declared glob(s)」; default tier taken; the seat's contract-tier review at delivery
Clause-②: no
Thread-read: the card body (filed by the triage seat with its labels; no grading comment on the thread at 2026-09-17T21:00Z)
Serial constraints cleared: PR #18756 (#18680, markerMatches) on this file LANDED 95e174527 at 2026-09-17T19:45:33Z; 30 open PRs' file lists read at 2026-09-17T20:54Z — none touches scripts/pm/check-half-states.mjs; os-verify-lock.sh --status at 2026-09-17T20:54Z: free; shared checkout at origin/main d9ba33df4; objectui's copy of this file is byte-pinned and re-synced only through objectui#9395, ⛔ not hand-mirrored. Hot-file queue behind this card: none. Slot: the second of the two freed by #18661's and #18693's deliveries (PR #18784 under the seat's review; PR #18786 in the merge queue); #18683 and #18527 in flight. Total order at 2026-09-17T21:00Z: p2 — #18536 (SKILL.md serial) · #18699 (behind PR #18725) · #18743 (SKILL.md serial) · #18744 (behind PR #18775; its deliverable burns the shared credential's GraphQL pool — the seat schedules it for a window with no landing pending) · #18755 (SKILL.md serial) · #18764 (behind #18683); p3 by age — #18622 · #18662 (behind PR #18784) · this card · #18665 (SKILL.md serial) · #18686 (behind #18744) · #18705 (MANDATORY, published skill) · #18721 (FREE, next).
Ruling-ref: the card body

---

Generated by Claude Code

[os-justin]

os-dev-report

{
  "issue": 18664,
  "status": "done",
  "branch": "claude/issue-18664-issue-body-limit-measured",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18796",
  "head_sha": "5fc8e0ce57374f90fe92032cc02017fe5c18c146",
  "files": [
    "scripts/pm/check-half-states.mjs"
  ],
  "session": "session_01Gqi43smmqjJ5sUrhfoPeKu",
  "premise_still_valid": true,
  "premise_notes": [
    "Artefact 1 MOVED since the card measured it: objectstack#6015 was 150,507 chars / 257,945 bytes for the triage seat; fetched 2026-09-17T21:03Z for this run it had been compacted to 72,754 chars / 125,561 bytes (updated_at 2026-09-17T20:04:42Z). The card's two numbers are quoted as the CARD'S, not re-taken. The falsification is re-taken independently and survives: 72,754 chars / 125,561 bytes is past 65,536 in EITHER unit.",
    "Artefact 2 confirmed on the tip this branch was cut from (6de7a2d6e6): ISSUE_BODY_LIMIT = 65536 at :17148 with the docblock 'GitHub's hard cap on an issue body ... A body that exceeds the cap is REJECTED by the API'; MARKDOWN_BODY_BUDGET = 60000 at :17149; the renderer cut at :19420; the pin at :28367.",
    "SEAT_BODY_SOFT_LIMIT is at :2037 on the tip (the card says :1802 — stale). Read, not touched, with H6.",
    "One nuance the card's causal sentence does not carry: what trims the report today is MARKDOWN_BODY_BUDGET at :19420, not ISSUE_BODY_LIMIT — the cap is what the budget was chosen to sit under, and it appears in the pins. Same defect, one extra step in the mechanism."
  ],
  "summary": "ISSUE_BODY_LIMIT is re-declared from a measurement, in the unit the platform actually refuses in. The refusal boundary was bisected to a single value on a throwaway issue opened for it (objectstack#18793): 262,144 bytes stored, 262,145 refused — 256 KiB exactly, bracketed on both sides. The unit was measured, not assumed: 262,145 bytes / 222,145 characters was refused where a character-counted cap would have had 40,000 to spare, and the same multi-byte shape at 262,144 bytes / 222,144 characters landed. The constant's docblock now carries unit, both sides of the bracket, the date and the probe issue. The renderer's trim at :19420 counted JS .length (UTF-16 code units) against a budget whose cap is in bytes, over rows that are largely CJK prose; it and familyLedgerReservation now count bytes, as H6 next door always has. MARKDOWN_BODY_BUDGET does NOT move (60,000, now bytes): bytes are never fewer than characters, so the new guard cuts at or before the old one — strictly narrowing, no widening of what the renderer prints. SEAT_BODY_SOFT_LIMIT and H6 untouched; objectui's byte-pinned copy not hand-mirrored.",
  "measurement": {
    "probe_issue": 18793,
    "probe_issue_url": "https://github.com/objectstack-ai/objectstack/issues/18793",
    "probe_issue_final_state": "closed, state_reason=completed, last landed body = the bisection record (2,960 bytes)",
    "write_budget_set": "18 body writes (the dispatch offered a dozen as a guide); 17 spent on measurement + 1 on the record",
    "tool": "scripts/pm/post-stamped.mjs --body=18793 (exit 0 + class identical = every byte stored; exit 4 EXIT_NOT_STORED = platform kept the old body)",
    "bisection_table": [
      "  # | size sent (bytes) |   chars | exit | class      | verdict",
      "  1 |           257,945 | 257,945 |    0 | identical  | landed",
      "  2 |           263,533 | 263,533 |    4 | not stored | refused",
      "  3 |           260,739 | 260,739 |    0 | identical  | landed",
      "  4 |           262,136 | 262,136 |    0 | identical  | landed",
      "  5 |           262,834 | 262,834 |    4 | not stored | refused",
      "  6 |           262,485 | 262,485 |    4 | not stored | refused",
      "  7 |           262,310 | 262,310 |    4 | not stored | refused",
      "  8 |           262,223 | 262,223 |    4 | not stored | refused",
      "  9 |           262,179 | 262,179 |    4 | not stored | refused",
      " 10 |           262,157 | 262,157 |    4 | not stored | refused",
      " 11 |           262,146 | 262,146 |    4 | not stored | refused",
      " 12 |           262,141 | 262,141 |    0 | identical  | landed",
      " 13 |           262,143 | 262,143 |    0 | identical  | landed",
      " 14 |           262,144 | 262,144 |    0 | identical  | LANDED  (the cap)",
      " 15 |           262,145 | 262,145 |    4 | not stored | REFUSED (one byte more)",
      " 16 |           262,145 | 222,145 |    4 | not stored | refused (multi-byte: the unit probe)",
      " 17 |           262,144 | 222,144 |    0 | identical  | landed  (multi-byte: the control)"
    ],
    "interval_reached": "A SINGLE VALUE, not an interval: max stored 262,144 bytes, min refused 262,145 bytes. Rows 1 and 2 re-take the card's two endpoints as this run's own bracket rather than inheriting them; rows 3-15 bisect. The card forbade calling 262,144 known unless exactly that size landed and one byte more was refused — rows 14 and 15 are that pair.",
    "unit_answer": "UTF-8 BYTES. Row 16 sent 262,145 bytes carrying only 222,145 characters (a 20,000-char run of U+4E2D plus ASCII); a cap counted in characters, or in UTF-16 code units (the same number here), would have taken it with 40,000 to spare. It was REFUSED. Row 17 is the positive control: the same multi-byte shape at 262,144 bytes / 222,144 characters LANDED, so the refusal is the one extra byte and not the content.",
    "incidental": "Every landed write read back class `identical` — no trailing-newline strip (these bodies carry none) and no synthesised footer on an ISSUE body. The cell post-stamped calls unmeasured (a body that already ENDS in the footer block) stays unmeasured; none of these did."
  },
  "constant_before_after": {
    "before": "export const ISSUE_BODY_LIMIT = 65536;  // docblock: \"GitHub's hard cap on an issue body ... A body that exceeds the cap is REJECTED by the API\" — no unit, no reading, no date",
    "after": "export const ISSUE_BODY_LIMIT = 262144;  // docblock carries UNIT (UTF-8 bytes), BOTH sides of the bracket (262,144 STORED / 262,145 REFUSED), DATE (2026-09-17) and PROBE ISSUE (objectstack#18793), plus the correction that the refusal is SILENT (old body kept, HTTP 200) rather than an API rejection, and the note that the round number is not the reason",
    "budget": "MARKDOWN_BODY_BUDGET unchanged at 60000, now explicitly BYTES. Margin stated in the docblock and pinned: 202,144 bytes below the cap, 22.9% of it. Raising it is a fold decision nobody has taken and is deliberately NOT taken here."
  },
  "guard_unit": {
    "before": "CHARACTERS — JS `.length`, i.e. UTF-16 code units: `used = body.length + indexText.length + ledgerReservation` and `used + line.length + 1 + notice.length > MARKDOWN_BODY_BUDGET`, over rows that are largely CJK prose, where bytes are 3x characters.",
    "after": "BYTES — a new tiny export `bodyBytes(text) = Buffer.byteLength(String(text ?? \"\"), \"utf8\")` in every term. The `+ 1` stays: the joining newline is one byte in UTF-8. `familyLedgerReservation` is reserved out of the same budget and now returns bytes too, which is what its own header has said since it was written; its two digit-slack terms are ASCII digits, already byte-correct. H6 next door (h6SeatBodyOversized) has counted Buffer.byteLength since it was written — the renderer now agrees with it and with the platform.",
    "direction": "Strictly NARROWING. Bytes are never fewer than characters, so the byte guard cuts at or before where the character guard did. No widening of what the renderer prints."
  },
  "pins": {
    "repointed": "The pin at :28367 now asserts bodyBytes(trimmed) is at most ISSUE_BODY_LIMIT, against the measured cap; twelve other size pins across the suite move to bytes with it (they asserted .length against a budget the guard no longer counts that way), plus the ledger-reservation and worst-case-ceiling measurements.",
    "new_battery": "SELF_TEST_BATTERIES gains 'ISSUE_BODY_LIMIT measured cap': 37 and SELF_TEST_BATTERY_FLOOR rises 5 -> 6 (the two sibling floor pins that name the roster size updated with it), in the shape PR #18756 landed. 40 cases register: 7 provenance (the docblock ATTACHED to the constant names unit, both bracket sides, date and probe issue — anchored on the docblock slice, not the file text where the assertion strings would satisfy themselves, with two controls that the extractor found a real docblock and not the whole file); 4 cap; 4 unit; 7 guard with its firing controls INSIDE it (a 4,000-row CJK report trimmed to the byte budget and announcing it, a non-vacuity pin that the row text really is 3x, and the COUNTERFACTUAL computed from the same run — how many rows a character-counting guard would have laid out and what that body would have weighed); 6 over-cap (the #4690 shape against the measured cap: a fixture whose rows alone exceed 262,144 bytes untrimmed, ASSERTED so it is not vacuous, renders under the cap AND the budget, SAYS it trimmed, points at the run log, never reaches the loud row); 4 margin; 6 floor.",
    "honesty_note": "The counterfactual overruns the renderer BUDGET, not the platform cap, which no character count can reach at today's budget — the battery comment says so in those words. The budget is what stops meaning anything in the wrong unit, and what would become dangerous the day anybody raises it toward the cap."
  },
  "ablation": {
    "from": "the COMMITTED fix 5fc8e0ce57; HEAD blob 153d10a0158ba27906a5400df8a9c486d77ae061",
    "method": "mutate on disk -> prove by grep -c on BOTH the removed and the injected text AND by git hash-object vs the HEAD blob (empty hash read as FAILURE) -> run the suite -> restore with `git checkout HEAD -- path` verified by hash equality and an EMPTY `git diff HEAD`, under `trap ... EXIT INT TERM` with absolute paths from `git rev-parse --show-toplevel`",
    "leg_a": "put 65536 back. Mutated blob 7d34ca5db0377d33d7e5062792dd54a51edbe763 (removed-text 1->0, injected-text 0->1). Self-test exit 1, 5 cases red, EVERY ONE in the new battery, nothing pre-existing red: the cap value (got 65536 want 262144), the 256-KiB identity, the retired-65,536 negative, the four-times relation, and the margin (got 5536 want 202144).",
    "leg_b": "put the character-counting accumulator back. Mutated blob d1ec008854fb628f8822232cbfa4344ace4d2342. Self-test exit 1, 9 cases red: 2 in the new battery (the CJK trim, the over-cap body's budget bound) and 7 PRE-EXISTING budget pins this PR re-measured in bytes (markdown budget x2, H17 budget, the instruction-4 box, #13947 order, #13947 reserved, #13947 flood). Reported as measured, not as a template default: those 7 could not fail under `.length`, and the reason they fail now is that this file's own fixtures already carry enough multi-byte text (em dashes in the row messages) that a character-bounded body overruns the byte budget it was supposed to keep. The old guard was not bounding the thing that matters, on this suite's own inputs.",
    "restore": "both legs restored: hash matches the HEAD blob and `git diff HEAD` empty; working tree clean afterwards (git status --porcelain empty)."
  },
  "tests": "node scripts/pm/check-half-states.mjs --self-test :: exit 0 — verbatim verdict line: \"check-half-states self-test: 4921 cases pass. Batteries: H66 released queue card 183/172, H19 judged-set founding 37/34, H65 tier declaration spelling 46/42, H67 queued merged-delivery reading 152/142, H2/H47/H66 decorated ownership marker 98/94, ISSUE_BODY_LIMIT measured cap 40/37.\" (4,881 cases before, 4,921 after.) Repo-wide `pnpm lint` (eslint . --no-inline-config) :: exit 0 at 5fc8e0ce57, whole repo, NO narrowing, so no narrowing declaration is owed. Control-character scan `grep -naP` over the C0 set plus DEL on the edited file :: exit 1, zero matches. Ablation as recorded above: rebuild not applicable (a .mjs run directly, no dist), on-disk mutation proved by grep counts on both texts plus blob-hash inequality, restored by hash under a trap.",
  "derived_gates": {
    "derivation": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack, run FROM the worktree with no hand-fed path list; it took its own change set from the merge base 6de7a2d6e (1 path: scripts/pm/check-half-states.mjs). 38 commands. Reconciled with --ran: 38 derived, 38 run, 0 unrun.",
    "all_37_fast_gates_exit_0": [
      "node scripts/check-changeset-no-major.mjs --base origin/main :: 0",
      "node scripts/check-changeset-no-major.mjs --self-test :: 0",
      "node scripts/check-ci-filter-parity.mjs :: 0",
      "node scripts/check-closing-keyword-parity.mjs :: 0",
      "node scripts/check-closing-keyword-parity.mjs --self-test :: 0",
      "node scripts/check-comment-mask-corpus.mjs :: 0",
      "node scripts/check-declaration-mirrors.mjs :: 0",
      "node scripts/check-declaration-mirrors.mjs --self-test :: 0",
      "node scripts/check-scripts-symbol-anchors.mjs :: 0",
      "node scripts/check-scripts-symbol-anchors.mjs --self-test :: 0",
      "node scripts/check-self-test-wired.mjs :: 0",
      "node scripts/check-self-test-wired.mjs --self-test :: 0",
      "node scripts/check-self-test-workflow-commands.mjs :: 0",
      "node scripts/check-self-test-workflow-commands.mjs --self-test :: 0",
      "node scripts/check-whole-set-label-write.mjs :: 0",
      "node scripts/check-whole-set-label-write.mjs --self-test :: 0",
      "node scripts/pm/bare-root-worklist.mjs --self-test :: 0",
      "node scripts/pm/board-snapshot.mjs --self-test :: 0",
      "node scripts/pm/check-governed-queue-guard.mjs --self-test :: 0",
      "node scripts/pm/sweep-closed-cards.mjs --self-test :: 0",
      "pnpm check:agent-test-spelling :: 0",
      "pnpm check:bash32-floor :: 0",
      "pnpm check:changeset-gate-self-tests :: 0",
      "pnpm check:cli-command-ids :: 0",
      "pnpm check:closing-target-claim :: 0",
      "pnpm check:commit-card-trailers :: 0",
      "pnpm check:cross-package-test-inputs :: 0",
      "pnpm check:driver-memory-census :: 0",
      "pnpm check:entry-guard :: 0",
      "pnpm check:nul-bytes :: 0",
      "pnpm check:parse-guard :: 0",
      "pnpm check:partof-closing-keyword :: 0",
      "pnpm check:pm-half-states :: 0",
      "pnpm check:pnpm-filter-targets :: 0",
      "pnpm check:ratchet-remedy-authority :: 0",
      "pnpm check:refd-timer-probe :: 0",
      "pnpm check:watch-hint-literal :: 0"
    ],
    "long_gate": "pnpm check:pm-dispatch-gates :: 0 — it runs past this container's foreground ceiling (745.9s on this box), so it was started DETACHED with output redirected to a file, never under a foreground timeout, and waited on with a foreground blocking wait. Verbatim verdict line: \"dispatch-gates self-test: 1788 cases pass.\"",
    "not_measured_by_design": "The derivation itself names what is OUTSIDE the 38: 48 artifact-roster families (silence is a fact about a LIST, not a clearance — 8 of them keep that roster under scripts/ or scripts/pm/, which this diff is in), 11 declared WIDE-population families, 10 pending-changeset families, 1 CI-measured-only family, 5 families taking a value from the workflow (including `check-half-states.mjs --format=markdown --provenance=...` itself), and 1 path-scheduled CI job with 5 steps. Those are CI's, not this run's, and their absence here is not a green."
  },
  "ci_state": "in_progress — the PR was opened at 5fc8e0ce57 and CI had not converged when this report was written. Waiting for CI is the PM seat's, not this run's.",
  "mcp_calls": "0 — no MCP GitHub tool was called, read or write. Every GitHub act went through the REST proxy with curl and GITHUB_TOKEN, Content-Type: application/json on every write.",
  "api_writes": "24 REST proxy writes. 1x POST /repos/objectstack-ai/objectstack/issues (create probe 18793); 18x PATCH /repos/objectstack-ai/objectstack/issues/18793 via post-stamped --body (17 measurement writes + 1 record body); 1x PATCH /repos/objectstack-ai/objectstack/issues/18793 (close, state_reason=completed); 1x POST /repos/objectstack-ai/objectstack/pulls (draft); 1x POST /repos/objectstack-ai/objectstack/issues/18796/labels (skip-changeset, read back: present); 1x POST /repos/objectstack-ai/objectstack/issues/18664/comments (this report). 1x PR body write, at creation only, never PATCHed. Plus 2 git pushes (not REST): the empty-branch write probe and the fix.",
  "open_questions": [],
  "out_of_scope_findings": [
    "to file (3 classes, dedupe words `65536 comment limit workflow log tail truncation`): five workflow files bound a log tail or a report against \"GitHub's 65536-character COMMENT limit\" — .github/workflows/cross-repo-issue-closer.yml:166, docs-drift-check.yml:557, merge-queue-triage.yml:178, scheduled-full-run-card.yml:125, test-nightly-tiers.yml:436. This run measured the ISSUE BODY cap only, so nothing here shows those five are wrong; what it shows is that they carry a number of the same provenance as the one just falsified, in a unit (\"character\") the body cap has now been measured NOT to use. Wants a card and its own bisection, on a file surface this card does not own.",
    "noted, not filed: FAMILY_LEDGER_WORST_CASE_BYTES's two measured figures (2,285 B / 3,754 B) were taken with .length under a name that says BYTES. Their pins now measure bytes and still fit the 6,000 ceiling, so nothing is broken — the prose numbers are simply readings in the other unit, and the docblock now says so. Successor: whoever next re-measures that ceiling; none queued.",
    "noted, not filed: raising MARKDOWN_BODY_BUDGET toward the measured cap is now possible and is deliberately NOT done here — it changes what the standing patrol's anchor issue prints, which is a fold decision for a seat or the maintainer, not a consequence of correcting a measurement. The dispatch forbids widening what the renderer prints. Successor: none — it wants a card of its own if anybody wants the longer report.",
    "noted, not filed: the card's causal sentence (\"按一个小 4 倍的假上限裁剪\") is one step off the code — the trim is against MARKDOWN_BODY_BUDGET, and ISSUE_BODY_LIMIT is the cap that budget was chosen to sit under and the value the pins assert. Same defect, extra step; recorded in the PR body so the next reader is not surprised. Successor: none needed."
  ]
}

---

Generated by Claude Code

[os-justin]

ACCEPT — in-seat contract-tier review of PR #18796 (head 5fc8e0ce5, report 5721642914) by the skills seat, 2026-09-17T21:52Z. Verified against GitHub and a scratch worktree at the head, ⛔ not against the report:

• Scope = the claim (5721158813), one stale number disclosed: one file, +227 / −31 in one commit — scripts/pm/check-half-states.mjs; skip-changeset correct (scripts/pm/** ships in no package); Clause-②: no on the PR body and the claim, check-clause2-carriers --pair 18796 0 (seat's run). No governed file touched; SEAT_BODY_SOFT_LIMIT (:2037 on the tip — the card's :1802 and the claim's copy of it were stale) and H6 untouched; objectui's byte-pinned copy not hand-mirrored. The card's causal sentence is one step off the code and the PR says so: the trim is against MARKDOWN_BODY_BUDGET, and ISSUE_BODY_LIMIT is the cap that budget sits under and the pins assert.
• The measurement, accepted as read and checked at its record: the refusal boundary bisected on a throwaway issue the dev opened and closed completed (objectstack#18793 — the seat read it: closed, completed, last landed body the 2,960-byte bisection record, opened by this seat's account), 17 measurement writes through post-stamped --body (exit 0 + class identical = stored; exit 4 = the old body kept): 262,144 bytes STORED, 262,145 bytes REFUSED — one value, bracketed on both sides, the card's own condition for calling 256 KiB known met by rows 14 / 15; rows 1 / 2 re-took the card's two endpoints rather than inheriting them. The unit is UTF-8 BYTES, measured: 262,145 bytes carrying only 222,145 characters was refused where a character cap would have taken it with 40,000 to spare, and the same multi-byte shape at 262,144 bytes / 222,144 characters landed. The card's artefact 1 ([PM seat] triage (objectstack-wide) — ⚪ 座位空缺 · 前任 os-sam · session_01U6nJJSVZb476ckbRxmaMCA 执 R+250→R+319 于 2026-09-18T21:0xZ 奉维护者令收班 · 收班简报 5736249665 · 定时器已停用(⛔ 勿复活,它绑旧会话)· 交接台账在正文 · objectstack 残差 20 张逐张点名 / objectui 0 · 接手先读必读 43(Routine prompt 带错规则) #6015) had been compacted before the dev read it (72,754 chars / 125,561 bytes at 21:03Z) — still past 65,536 in either unit, so the falsification survives its re-take; disclosed.
• The fix, read in the diff: ISSUE_BODY_LIMIT = 262144 with a docblock carrying unit, both bracket sides, date and probe issue, and the correction that the refusal is SILENT (old body kept, HTTP 200) rather than an API rejection; a new export bodyBytes(text) = Buffer.byteLength(…, 'utf8'); renderMarkdown's accumulator and cut at :19481+ and familyLedgerReservation now count BYTES (the + 1 stays: one newline byte) — the old guard counted UTF-16 code units against a byte budget over rows that are largely CJK prose; MARKDOWN_BODY_BUDGET stays 60,000, now explicitly bytes, with the margin stated (202,144 B below the cap, 22.9 %) and NOT raised — a fold decision deliberately not taken. Direction: strictly NARROWING (bytes are never fewer than characters), no widening of what the renderer prints.
• Seat's own readings on the head: ISSUE_BODY_LIMIT 262144, MARKDOWN_BODY_BUDGET 60000, bodyBytes('中') 3, batteries 6 / floor 6.
• Pins: a new battery ISSUE_BODY_LIMIT measured cap (40 registered, floor 37; SELF_TEST_BATTERY_FLOOR 5 → 6, the two sibling roster-size pins updated): 7 provenance pins anchored on the constant's own docblock slice (with two controls that the extractor found a real docblock and not the whole file), 4 cap, 4 unit, 7 guard with its firing controls INSIDE it (a 4,000-row CJK report trimmed to the byte budget and announcing it, a non-vacuity pin that the rows really are 3×, the COUNTERFACTUAL from the same run — the rows and bytes a character-counting guard would have laid out), 6 over-cap (the check:react-declaration-parity 是唯一没接进任何 workflow 的源码审计门禁,且无 MANIFEST 时静默 skip 退出 0 —— 它现在永远不可能红 #4690 shape against the measured cap: a fixture whose rows alone exceed 262,144 bytes untrimmed, asserted non-vacuous, renders under cap AND budget, says it trimmed, points at the run log), 4 margin, 6 floor; :28367 re-pointed to bodyBytes(trimmed) ≤ ISSUE_BODY_LIMIT, twelve size pins across the suite moved to bytes with it. The honesty note is on the record: the counterfactual overruns the BUDGET, not the platform cap.
• Ablation (accepted as read): two legs from the committed fix, each proven on disk by grep counts on both texts and blob inequality (153d10a01 → 7d34ca5db / d1ec00885), restored by hash and an empty git diff HEAD under a trap — leg A (65536 back): 5 red, all in the new battery, nothing pre-existing; leg B (the character accumulator back): 9 red — 2 in the new battery and 7 PRE-EXISTING budget pins that could not fail under .length and fail now because the suite's own fixtures carry enough multi-byte text to overrun a byte budget under a character bound — reported as measured, and it is the measurement that the old guard was not bounding the thing that matters.
• Self-test on the head, seat's own run: check-half-states.mjs --self-test exit 0 — 4921 cases (4881 on main).
• Gates: 38 derived, 38 run, all exit 0, check:pm-dispatch-gates DETACHED (1788 cases, 746 s) and waited on in the foreground, --ran 38 / 38; repo-wide pnpm lint exit 0; control-byte scan clean — per the report. Seat-side: the self-test and the constants above, pair 0, expected-skips 0, harness CURRENT. CI on the head: 37 check runs read explicitly, every one completed — 26 success, 11 skipped (all in the expected-skips roster, exit 0); commit status success; none failure.
• Writes: api_writes 24 — the probe issue created, 18 body writes to it (17 measurements + the record), its close, the draft PR, skip-changeset read back, this report; mcp_calls 0; PR body written once at create, never PATCHed. Within the budget the dispatch set (「a dozen is plenty」 was a guide; 17 measurements bracketed a single value, said so).
• Out-of-scope, carried: (1) TO FILE by the seat: five workflow files bound a log tail or report against 「GitHub's 65536-character COMMENT limit」 (cross-repo-issue-closer.yml :166, docs-drift-check.yml :557, merge-queue-triage.yml :178, scheduled-full-run-card.yml :125, test-nightly-tiers.yml :436) — same provenance as the number just falsified, in a unit the body cap was measured NOT to use; wants its own bisection on the COMMENT surface; (2) FAMILY_LEDGER_WORST_CASE_BYTES's two prose figures were taken in the other unit — docblock now says so; (3) raising the budget toward the cap is possible and deliberately not done — a fold decision for a card of its own.
• Landing: non-governed (scripts/pm/**) ⇒ this seat's ACCEPT is the review of record; ready + auto-merge SQUASH through the CCR route follow now (CI complete and green). Serial behind it on this file: none queued.

---

Generated by Claude Code

[os-justin]

LANDED — PR #18796 (#18664, ISSUE_BODY_LIMIT declared from a measurement, in the unit the platform refuses in) merged through the queue at 2026-09-17T22:15:51Z (merged_at), squash 890488060 (single parent 034f5a3af, an ancestor of origin/main; ISSUE_BODY_LIMIT = 262144 and bodyBytes() read in check-half-states.mjs on main, the renderer's accumulator, cut and familyLedgerReservation in bytes, MARKDOWN_BODY_BUDGET 60000 unchanged, its --self-test at this head answered 4921 cases); armed by this seat through the CCR route after the in-seat ACCEPT 5721727774 — ready 2026-09-17T21:52:19Z, added_to_merge_queue 2026-09-17T21:52:20Z. The card auto-closed completed at 2026-09-17T22:15:52Z on the PR's Fixes; pm:dispatched is stripped in the same act. Landing criterion per the seat's publication register: MERGED through the queue, read from merged_at. The measurement stands at its record (objectstack#18793, closed completed): 262,144 bytes STORED, 262,145 bytes REFUSED — the old 65536 was false in both directions and is gone from the script; objectui's byte-pinned copy is re-synced only through objectui#9395, ⛔ not hand-mirrored. Carried out of this landing by the seat: the five workflow files bound against 「GitHub's 65536-character COMMENT limit」 (to file as a finding, its own bisection on the COMMENT surface). Serial behind it on this file: none queued.

---

Generated by Claude Code