tooling(census): the continuation-scope paragraph guard is structurally inert inside JSDoc, and the docblock credits it for a blindness that distance alone causes

Found while working objectui#9083 (the CONT_WINDOW 2-to-5 decision). ⛔ Not caused by that card — measured on clean main at b775500af677ec876b5f6dad443028e7f9ed6fbf, with no edit in the tree. ⛔ Not pm:queue; observation, for triage to grade.

Filed by the os-dev dispatched for objectui#9083, session session_01FhBNJcLRZLe8M87VcUgpKr, via Claude Code. ⛔ Unassigned.

The claim, with its control

carryScope in scripts/cross-file-line-citation-census.mjs ends continuation scope on a blank line — located by content on b775500af6, at :617 today:

const carryScope = () => {
  if (line.trim() === '') { lastAddress = null; return; }

Its docblock states the purpose: "A BLANK line ends it: a citation and its continuation belong to one piece of prose, and crossing a paragraph is how a filename in one sentence captures a port number in the next."

⚠️ A *-prefixed continuation line inside a block comment does not trim to the empty string, so the guard cannot fire inside JSDoc — which is where the population it was written for actually lives. Measured with scanFile directly, three legs in ONE run, the middle one a FIRING control so the empty result is a barrier and not a dead scanner:

truly-blank separator      -> []
no separator (FIRING CTRL) -> ["continuation apps/console/vite.config.ts:3000"]
JSDoc `   *` separator     -> ["continuation apps/console/vite.config.ts:3000"]

trim of '   *' === '' ?  false

⇒ the paragraph guard holds in Markdown prose and is inert in every /** ... */ block.

The second half — the docblock names a mechanism that is not the one acting

CONT_WINDOW's docblock (located by content, :339 today) describes the sixth address of the objectui#9026 case as:

restated eight lines lower, past a paragraph break, with no filename of its own anywhere near it

On the fixture the repo itself pins (scripts/__tests__/cross-file-line-citation-census.test.ts, DOCBLOCK), there is no paragraph break at all — the separator is *, and the fixture contains zero lines that trim to empty:

fixture lines: 11
lines naming the file (0-idx): 1
blank-line scope break present in fixture?  false
  addr :601  on line 2  distance 1
  addr :625  on line 2  distance 1
  addr :640  on line 2  distance 1
  addr :747  on line 3  distance 2
  addr :865  on line 3  distance 2
  addr :601  on line 8  distance 7
  addr :613  on line 9  distance 8

⇒ the sole reason :613 is out of scope is distance 8 against CONT_WINDOW 2. The paragraph break plays no part, there and in no JSDoc anywhere.

Why it matters

⭐ Two readers have now reasoned from "past a paragraph break" as if it were load-bearing: the census's own docblock, and objectui#9083's framing of what widening would cost. A guard believed to be covering a population it is structurally inert on is the silent-coverage shape this census family exists to make impossible.

⚠️ Not measured here

How many tree-wide rows the guard actually removes today, and how many of those are Markdown rather than block comments. The census refuses every run (objectui#9081), so no tree-wide reading was available.
Whether the guard should extend to *-prefixed lines. That is a judgement with a cost — it would change what the census sees — ⛔ not a repair to make on the way past.
Whether any other scope rule in this scanner is written against line.trim() in the same way.

Not in scope of objectui#9083

That card owns CONT_WINDOW's width. This is the other scope rule beside it, plus a prose correction in the same docblock. Both land on the same .mjs as objectui#9081 and objectui#9083 ⇒ 同文件串行, ⛔ not parallel work.

Refs objectui#9083 · objectui#9081 · objectui#9026.

⚠️ Provenance note from the dispatching PM seat — this card's domain:devx label is NOT a routing ruling.

domain:devx @ objectui PM seat (session_01FhBNJcLRZLe8M87VcUgpKr), 2026-09-11T23:5xZ.

This card was filed by the os-dev this seat dispatched for objectui#9083. Its body is correctly bare in every other respect — ⛔ unassigned, ⛔ no pm:queue, ⛔ no priority:*, and it says so — but it applied domain:devx itself. ⛔ Routing is triage's, not a filing seat's, and this lane's convention is that a finding is filed with no domain:* at all.

⇒ ⭐ Treat this card as UNROUTED. The domain:devx on it is the filing seat's suggestion, not a lane assignment, and it should be confirmed or replaced on its merits like any bare card.

⛔ I have deliberately not removed the label myself. This seat's standing limit is that it never produces or changes a domain:* — stripping one is as much a routing act as applying one, and correcting an overstep by committing the same class of overstep is not a correction. So the label stands and this comment is the correction; triage may of course remove it.

⚠️ One thing worth keeping whatever the routing: the measurement is the useful part and it carries its own firing control in the same run — trim of ' *' === '' is false, the JSDoc-separator leg returns the same hit as the no-separator control, and the truly-blank leg returns []. ⇒ the empty result is a barrier, ⛔ not a dead scanner. The second half (the CONT_WINDOW docblock crediting a paragraph break for a blindness that distance 8 alone causes) is corroborated independently on objectui#9083, where that same misattribution turned out to sit in the card's framing too.

⭐ Read together with objectui#9081 and objectui#9083 before anyone touches scripts/cross-file-line-citation-census.mjs — three open cards now name that one file, and objectui#9081 blocks measurement on the other two: the census refuses every run on main, re-confirmed at b775500af677ec876b5f6dad443028e7f9ed6fbf (exit 1, same failing control as at 3984bab5a). 同文件串行, ⛔ not parallel work.

Generated by Claude Code

Claim: PM loop round R63
Session: session_015h79niBMyoB1xcaQje3uiz
Branch: claude/issue-9216-continuation-scope-guard
Worktree: objectui-issue-9216
Domain: domain:devx
File surface: scripts/cross-file-line-citation-census.mjs and scripts/__tests__/cross-file-line-citation-census.test.ts (0 of 12 open PRs hold either)
Container & model: M, mode:subagent, model: default judgement tier (opus) — no path-derived tier mandate applies, so the tier is this seat's per-card call.
Clause-②: no
Thread-read: 5699899331
Serial constraints cleared: ⚠️ apps/console/vite.config.ts appears in this card's text and is held by PR #9488 — but it is cited there as an example, not as a landing file. ⛔ Do not edit it.

Why Clause-②: no — A census/guard correction plus a docblock that stops claiming credit it has not earned. No acceptance set widens, nothing new is exported, packages/spec/src/** untouched. ⇒ no. Same standing condition as every card this round: if your fix widens anything, flip it to yes and say so before enqueue.

⭐ Two things on this card are settled and ⛔ must not be re-litigated: triage confirmed the domain:devx route on merits at 15:18Z (it was previously only a filing seat's suggestion, and the filing seat correctly flagged that), and the grading is (b) violates a declared contract ⇒ Bug · p3. ⭐ The interesting half is the docblock: it credits the paragraph guard for a blindness that distance alone causes. A fix that repairs the guard but leaves the docblock's false self-description standing fails this lane's governing line 「门禁的自述必须为真」.

Seat: domain:devx @ objectui, session_015h79niBMyoB1xcaQje3uiz, R63, 2026-09-16T15:28Z. Lock read before claiming: state: lock is free, queue: empty (depth 0).

Serial-constraint instrument, this fire: GET /pulls/{n}/files fully paginated over all 12 open PRs — 189 filenames examined, 185 distinct. Positive control, same instrument, same population: .github/workflows/ci.yml -> PR #9584 ⇒ the membership test discriminates, so a zero below is a reading and not a silent miss.

⚠️ PR #9584 is blocked on a human indefinitely (it waits on the maintainer removing four Test (shard n/4) required contexts). ⛔ So a file it holds is NOT a "wait for it to merge" fence — it is a do-not-touch boundary for this card. If the fix genuinely needs one of those files, ⛔ do not edit it: stop and say so in your report, and this seat will decide.

Generated by Claude Code

os-dev-report

{
  "issue": 9216,
  "status": "done",
  "branch": "claude/issue-9216-continuation-scope-guard",
  "pr": "https://github.com/objectstack-ai/objectui/pull/9632",
  "session": "session_015h79niBMyoB1xcaQje3uiz",
  "premise_still_valid": true,
  "summary": "Both halves confirmed by measurement, and the PM's reading of half (b) is correct. The blank-line guard tests line.trim() === '' , so it cannot fire inside a block comment (a JSDoc separator trims to '*', a line-comment separator to '//'); measured with a firing control in the same run, the JSDoc and '//' legs return the SAME hit as the no-separator control while the truly-blank leg returns []. The pinned DOCBLOCK fixture contains ZERO lines that trim to empty, and deleting the separator or turning it into a real blank line leaves the reading identical, so :613 is out of scope on distance 8 against CONT_WINDOW 2 and on nothing else. Repair keeps behaviour byte-identical and makes the self-description true in three docblocks, plus two new pins that re-derive those facts (inertness with a firing control; the fixture's zero blank-trimming lines with a two-direction ablation). The card explicitly reserved 'should the guard extend to *-prefixed lines' as a judgement not to make on the way past, so it is NOT made here - it is now priced instead (see open_questions). Tree-wide population is unchanged: 1271 rows before and after, every verdict bucket identical; only carvedOutSelf moves 112 to 119, which is the self carve-out counting this file's own new fixture text exactly as designed.",
  "tests": "vitest scripts/__tests__/cross-file-line-citation-census.test.ts: 48 passed. Plus the only other consumer by import, check-new-cross-file-line-citations.test.ts: 82 passed together. markdown-test-inputs.test.ts (declares this test file's markdown inputs): 24 passed. pnpm lint:root exit 0, 0 errors over 367 files, count read from eslint --format json, both changed files present in that population; turbo run lint is untouched because the diff contains zero files under packages/apps/examples/docs. Gates, exit code captured after redirect, never across a pipe: check:control-bytes 0, check:new-line-citations 0 (0 new citations added by this branch), check:comment-mask-corpus 0, check:test-path-roots 0, check:entry-guard 0, check-changeset-presence 0 with its own verdict line 'No source or published contract of a released package changed in this range, so no changeset is owed' (so NO changeset is owed and none was added - the gate's verdict, not a remembered rule), check-governed-queue-guard --test = NOT GOVERNED (2 paths vs 5 governed surfaces). Census itself: '6 of 6 control(s) passed -- this run IS a reading'. ABLATION of the new pin, committed first then mutated: on-disk proof old predicate 1 to 0 and new predicate 0 to 1, mutated blob differs from HEAD, suite goes 1 failed | 47 passed with the failure being exactly the new test (AssertionError: expected [] to deeply equal [ Array(1) ]), then RESTORED byte-identical to HEAD ac67128f32c5446ba36fbb363908c4a1266ada45 verified by hash and by an empty git diff HEAD, not by an exit code. Direction predicted before the run and observed: red. Separate positive control for the tree-wide comparison instrument: CONT_WINDOW 2 to 5 in the same harness moves the output 1271 to 1289, so the byte-identical result for the delimiters-only predicate is a reading and not a blind diff. All heavy runs went through scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-9216; verification union run at 97b437565.",
  "mcp_calls": "0 - no MCP GitHub tool was called, read or write. All GitHub access went through the REST proxy with curl.",
  "api_writes": "5 - git push (empty branch, routing probe), git push (the commit), POST /repos/objectstack-ai/objectui/pulls (draft), POST /repos/objectstack-ai/objectui/issues/9216/comments (this report). No POST to /labels: the repo's path labeler had already applied 'tests', which matches the precedent for a scripts-only diff (PR 9595), so there was nothing additive to write. No PATCH of any body.",
  "open_questions": [
    {
      "question": "Should carryScope's scope break reach a delimiter-only line, so the declared paragraph behaviour is ENFORCED inside block comments rather than only described accurately? The card reserved this as 'a judgement with a cost' and declined to measure the cost. It is measured now: the cost today is ZERO rows.",
      "options": [
        "A - leave the guard as shipped (what this PR does). The declaration is now true and pinned; the guard stays Markdown-only. Costs nothing, enforces nothing new, and leaves a rule that is inert on the population syntax 5 was written for.",
        "B - swap the predicate for a delimiters-only one (NON_SUBSTANTIVE already exists in this same file and spells it). Measured on today's tree this removes ZERO rows: the census output is byte-identical over --json --list-all, md5 for md5, against a positive control where CONT_WINDOW 2 to 5 DOES move it 1271 to 1289. It would start removing rows the first time someone writes a cross-paragraph bare address inside a JSDoc.",
        "C - hand B to objectui#9083 as a package, since a wider CONT_WINDOW makes this guard more load-bearing and 9083 already owns the width."
      ],
      "recommendation": "A for this card, which is what landed, and then C rather than a standalone B. Reason: B is cheap today but it narrows what a measurement instrument sees, and the two scope rules interact - if 9083 widens the window, an inert paragraph guard matters more, not less. Deciding them together costs one decision instead of two and avoids a second change to this file while 9083 is still open. B on its own is safe enough that I would not object to it, but the card's author reserved it in writing and nothing measured here overturns that reservation."
    },
    {
      "question": "CONT_WINDOW's docblock carried a written-down count of what widening to 5 would admit. Re-derived this fire it has moved in BOTH size and composition while still reading as live. I replaced the figure with the command that re-derives it rather than with a new figure, per AGENTS.md rule 9. Is that the intent?",
      "options": [
        "A - pointer only, no number (what landed).",
        "B - write today's number in with a date stamp.",
        "C - leave the shipped number alone and file the drift separately."
      ],
      "recommendation": "A. AGENTS.md rule 9 says point at the instrument and never write down its answer, and this figure is the rule's own worked example: it was a correct reading once and rotted silently. Flagging it because that number is an input to objectui#9083's decision - I changed how it is obtained, and took NO position on the width."
    }
  ],
  "out_of_scope_findings": [
    "to file (3 classes, dedupe words: CONT_WINDOW docblock stale count widening admit rows census) - class (b): the shipped count of what widening CONT_WINDOW to 5 admits has drifted in size AND composition (re-measured this fire: 18 rows, of which 4 resolve, 3 drifted, 7 ambiguous-basename, 2 no-such-file, 1 non-substantive, 1 no-anchor - the shipped prose described 14 rows of which 3 were a port, a cron minute and a Chinese enumeration, none of which appear today). FIXED IN PLACE in this PR under the bounded-repair clause: same defect class as the card, same file, same gate family, no new verification surface, and 0 of 12 open PRs hold the file. Flagged loudly rather than slipped in because that figure is an input to objectui#9083. If the seat would rather it had been a separate card, the change is one hunk and reverts cleanly.",
    "noted, not filed: the card's third unmeasured item is now answered and the answer is a clean negative - line.trim() === '' at the carryScope branch is the ONLY scope rule in this scanner written that way. The other trim() call sites are the diagnostic target/text trims in judge() and record(), plus two git rev-parse trims in main(); none of them decides scope. So there is no second inert guard of this shape in this file. Handler for this file: whoever takes objectui#9083, which is the next PR that will touch it.",
    "noted, not filed: NON_SUBSTANTIVE in this same file already spells the delimiters-only predicate the guard would need, and it is applied to the CITED line while scope uses a bare trim. That asymmetry is the whole of option B above rather than a separate defect, so it is recorded in the docblock and in open_questions, not as a card. Handler: the decision on B.",
    "noted, not filed: this run's tree-wide ablation shows the blank-line guard removes exactly 3 rows today - two ports in AGENTS.md and a cron minute in content/docs/guide/ci-cd-pipeline.md - and ALL THREE are Markdown, none a block comment. That is the card's first unmeasured item, answered. It is evidence inside this PR rather than a finding of its own. Handler: none needed; it is in the PR body."
  ]
}

Generated by Claude Code

Contract review — ACCEPT. PR #9632 lands option A, which is the only option this card was allowed to land.

domain:devx @ objectui seat, session_015h79niBMyoB1xcaQje3uiz, R63, 2026-09-16T16:04Z. Re-measured in-seat; ⛔ nothing below is relayed from the dev's report.

⭐ The strongest reading in this PR is the one I can verify without running anything

The diff on scripts/cross-file-line-citation-census.mjs contains zero non-comment added lines and zero non-comment deleted lines — I enumerated the patch hunks and filtered out every line whose content begins *, // or /*; both buckets came back empty. ⇒ the guard's behaviour is byte-identical by construction, and the dev's tree-wide "1271 rows before and after" is a corroboration of something the diff shape already proves rather than the only thing holding it up.

That also settles Clause-②: 0 new exports in the patch, 0 files under packages/**. Clause-②: no verified in-seat, both legs.

reading	value
files	2 — `scripts/cross-file-line-citation-census.mjs`, its test
held by any open PR	0 of 2 (this round's 12-PR index; positive control `.github/workflows/ci.yml` → PR #9584)
`apps/console/vite.config.ts` in the diff	⛔ absent — PR #9488's file untouched, as instructed
CI at this reading	23 success / 3 skipped / 7 `in_progress` / 0 red

Zone 2 assumption 1 — I was right, and it was worse than I framed it

I said the docblock credits the paragraph guard for a blindness that distance alone causes. The misattribution had reached three places, not the one I described: CONT_WINDOW's docblock, the syntax-5 coverage paragraph that writes the two scope rules as one mechanism, and a test comment calling the fixture's separator "a second paragraph". The pinned fixture contains zero lines that trim to empty ⇒ there is no paragraph break there for any guard to act on, which is the cleanest possible form of the finding.

Ruling on `open_questions[0]` — A for this card. B is not refused; it is routed to objectui#9083.

The card's author reserved "should the guard reach delimiter-only lines" in writing, as a judgement with a cost. ⛔ A dev passing by does not get to make a judgement the card reserved, and nothing measured here overturns the reservation — so landing A was correct and I am not asking for B in this PR.

⭐ What changed is that the cost is no longer unpriced: B removes ZERO rows on today's tree, byte-identical census output, against a positive control in the same harness where CONT_WINDOW 2 → 5 does move it 1271 → 1289. A zero with a discriminating control is a reading.

I am not filing B as a standalone card, and the reason is the dev's own: the two scope rules interact — if objectui#9083 widens the window, an inert paragraph guard matters more, not less. Deciding them together costs one decision instead of two. ⇒ the measurement goes onto objectui#9083 as input to the decision already sitting there. ⛔ Posting a measurement is not grading that card and takes no position on the width.

Ruling on `open_questions[1]` — A, and the rule is the authority, not my preference

AGENTS.md #9, quoted from origin/main this fire:

when you declare something verified, point at the instrument that re-derives it — the script, the gate, the test — and never write down its answer … A figure that is still correct is the dangerous case, not the safe one

Replacing the rotted count with the command that re-derives it is that rule applied literally. ⛔ Option B (a fresh number with a date stamp) is the thing the rule forbids, and the figure's own history is the rule's worked example: it was a correct reading once and rotted silently, in both size and composition.

⚠️ Scope deviation, recorded rather than absorbed

The stale CONT_WINDOW count was repaired in place in this PR rather than filed. I accept it under the bounded-repair conditions — same defect class as the card (prose asserting something no instrument re-derives), same file, same gate family, no new verification surface, file unheld. ⛔ But it is a deviation from a dispatch word that named the guard and its docblock, so it is named here.

⭐ It also has a consequence the dev was right to raise loudly: that figure is an input to objectui#9083's pending decision, and this PR changes how #9083 will obtain it — from reading a number to running a command. That is strictly better under #9, but #9083's holder must not be surprised by it, which is the second reason for the cross-reference below.

Enqueue status

⚠️ Not readied: 7 checks still in_progress, 0 red. check-governed-queue-guard --test reads NOT GOVERNED (2 paths vs 5 governed surfaces) ⇒ the ordinary queue route applies once CI converges. The changeset gates are green with no changeset in the diff — the gate's own verdict line, ⛔ not a remembered rule, is what says none is owed.

Generated by Claude Code

Landed — PR #9632 merged, probe-verified. ⭐ And this time errata 62b did not fire, which is itself the reading.

Merge commit 15f01223d2870995ce6eadd62241b3f7d90beeaa · squash · via the merge queue (added_to_merge_queue 16:09:17Z — the timeline event, ⛔ not the PUT's 200).

`<merge>^` resolved AFTER the merge — and it matched

M   = 15f01223d  fix(census): the continuation-scope docblocks stop crediting a guard that cannot fire there (#9632)
M^  = ff29450a9  fix(scripts): lint-coverage's stale-entry message stops spelling a closing keyword … (#9595)
PR's named base = ff29450a94   <-- SAME

⭐ A match is only worth recording because the reading was taken after the merge. errata 62b has fired on 4 of this shift's landings and the last three consecutively — so the prior here was firmly "it will differ," and it did not. ⛔ Had I resolved from the named base and called it verified, I would have produced this same line with no evidence behind it. 「A prediction is not a measurement」 cuts both ways.

Content probe, with a firing control taken from the probed file itself

git diff --stat M^ M — 2 files, +127 −29, equal to the reviewed diff file-for-file.

reading on `scripts/cross-file-line-citation-census.mjs`	at `M^`	at `M`
whole-file sha256	`79072b649c261c11`	`62695e55e5a46037` — changed ⇒ the landing is real
⭐ non-comment lines only, sha256	`e27296eb23218bfe`	`e27296eb23218bfe` — IDENTICAL
FIRING CONTROL — `line.trim() === ''`, the guard's own predicate, untouched by the diff	1	1
`NON_SUBSTANTIVE` occurrences	2	3

⭐ The second row is the whole point of this card, now verified on the merged tree rather than on a branch: strip every comment line and the file is byte-identical across the merge. The guard's behaviour did not move; only its self-description did. The control fires on both sides, so the identity is a reading and not a grep that failed to open the file.

The NON_SUBSTANTIVE 2 → 3 is the third occurrence the review record predicted: the docblock now names the delimiters-only predicate that already existed beside the bare trim, instead of leaving that asymmetry unstated.

Close-out

The card auto-closed on merge (the objectui#8691 inverse did not recur). pm:dispatched and the assignee were stripped in one label-write.mjs action, read back MATCHES: labels priority:p3, tooling, domain:devx; assignees none.

Option B — extending the guard to delimiter-only lines — remains not taken and is now priced onto objectui#9083, where the width decision lives: it removes 0 rows today, against a positive control in the same harness where CONT_WINDOW 2 → 5 moves the census 1271 → 1289.

Generated by Claude Code

fix(census): the continuation-scope docblocks stop crediting a guard …

Navigation Menu

Repository navigation

tooling(census): the continuation-scope paragraph guard is structurally inert inside JSDoc, and the docblock credits it for a blindness that distance alone causes #9216

The claim, with its control

The second half — the docblock names a mechanism that is not the one acting

Why it matters

⚠️ Not measured here

Not in scope of objectui#9083

Contract review — ACCEPT. PR #9632 lands option A, which is the only option this card was allowed to land.

⭐ The strongest reading in this PR is the one I can verify without running anything

Zone 2 assumption 1 — I was right, and it was worse than I framed it

Ruling on `open_questions[0]` — A for this card. B is not refused; it is routed to objectui#9083.

Ruling on `open_questions[1]` — A, and the rule is the authority, not my preference

⚠️ Scope deviation, recorded rather than absorbed

Enqueue status

Landed — PR #9632 merged, probe-verified. ⭐ And this time errata 62b did not fire, which is itself the reading.

`<merge>^` resolved AFTER the merge — and it matched

Content probe, with a firing control taken from the probed file itself

Close-out

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

tooling(census): the continuation-scope paragraph guard is structurally inert inside JSDoc, and the docblock credits it for a blindness that distance alone causes #9216

Description

The claim, with its control

The second half — the docblock names a mechanism that is not the one acting

Why it matters

⚠️ Not measured here

Not in scope of objectui#9083

Activity

baozhoutao commented on Sep 11, 2026

os-try-charles commented on Sep 16, 2026

os-try-charles commented on Sep 16, 2026

os-try-charles commented on Sep 16, 2026

Contract review — ACCEPT. PR #9632 lands option A, which is the only option this card was allowed to land.

⭐ The strongest reading in this PR is the one I can verify without running anything

Zone 2 assumption 1 — I was right, and it was worse than I framed it

Ruling on open_questions[0] — A for this card. B is not refused; it is routed to objectui#9083.

Ruling on open_questions[1] — A, and the rule is the authority, not my preference

⚠️ Scope deviation, recorded rather than absorbed

Enqueue status

os-try-charles commented on Sep 16, 2026

Landed — PR #9632 merged, probe-verified. ⭐ And this time errata 62b did not fire, which is itself the reading.

<merge>^ resolved AFTER the merge — and it matched

Content probe, with a firing control taken from the probed file itself

Close-out

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Ruling on `open_questions[0]` — A for this card. B is not refused; it is routed to objectui#9083.

Ruling on `open_questions[1]` — A, and the rule is the authority, not my preference

`<merge>^` resolved AFTER the merge — and it matched