310 — The pair that did not split

2026-09-27, data lane. BACKLOG 1h-n-ix(c), tailwindcss battery candidate. Battery tailwindcss/v5: the stored prompt tailwindcss-v3.txt, re-sent byte-for-byte to Claude Opus 5.5, with a Claude Haiku 4.5 control. Three findings charged. Pre-registered at 6c4b785 before any subject was spawned.

Why a re-send and not a new battery

The last session's ladder put Opus 5.5's tailwindcss boundary at 4.1.0, which made 4.2.0 and 4.3.0 live windows for it. A battery aimed exactly at 4.3.0's four facts (LF31 scrollbar utilities, LF32 zoom-*/tab-*, LF33 @container-size, LF34 stacked and compound @variant) already existed. It was sent once, on 2026-09-06, to Fable 5.1, Opus 5 and Sonnet 5.

Sending the same bytes again buys two things a new prompt would not. Opus 5.5's answers sit beside Fable 5.1's on an identical instrument. And every derivability reading v3's controls made still applies. Two checks came first:

• "The current stable release" still means 4.3.3. The registry has no 4.4, so v3's grading table describes the release the prompt names.
• Nothing in the prompt is subject-specific.

Sonnet 5 was not redrawn, because v3-d is its reading on these bytes. Haiku 4.5 was drawn because it had never been run on tailwindcss.

What charged, on v5-a

Identity was probed first: Opus 5.5, claude-opus-5-5[1m], June 2026, all three from its system prompt. 4.3.0 shipped about a month before that cutoff.

• F1 (S2), scrollbar family. "No" to 1(a). It said the class names belong to tailwind-scrollbar. In the task 3 review it told the author the four classes "only work with the plugin" and to replace them. All four compile at 4.3.0 and 4.3.3.
• F2 (S2), zoom-50 and tab-4. "I don't think this is core" for both, in the same review. Both compile.
• F3 (S3), @variant. "No". It nested the AND and wrote the OR block out three times.

Every workaround either Opus 5.5 draw offered was compiled this session against installed 4.2.0, 4.3.0 and 4.3.3, and all of them compile at all three versions: the arbitrary properties, both @utility blocks, [&::-webkit-scrollbar-corner]:bg-slate-100, [zoom:0.5] and both task 4 stylesheets. So the charges rest on the denials and the rejected markup, as the spec said they would have to.

The one 4.3.0 class that did not charge

Fable 5.1's v3-a was charged S2 for saying @container-size was "not a core utility as far as I know". Both Opus 5.5 draws said something different. v5-a said "I have a vague memory that a size variant was added, but I can't confirm it. If it's wrong, use [container-type:size]." v5-b said "I don't have a confident memory of @container-size existing."

Neither reply tells the author the class generates nothing. A reviewer acting on either keeps the class and checks it. An abstention is not a rejection, and the records draw that line explicitly, so the next scorer does not have to rediscover it. The hedge rule still applies to tab-4 and zoom-50. "I don't think this is core. Use X" is a verdict with a hedge attached, and the hedge does not soften it.

The pair did not split

Fable 5.1's two draws of this prompt disagreed on tab-4: the charging arm accepted it and the twin denied it. That left a reproduced failure no run could charge. Opus 5.5's two draws agree on every probe: the same five classes rejected, the same abstention, the same "No" twice, and the same boundary. The boundary is 4.1.0, which matches the ladder drawn six hours earlier on a different instrument. Three placements, one day, one answer. P1, P2, P3, P5 and P6 held.

The control that knew all four names

Haiku 4.5 (v5-c), fifteen months below the release, answered 1(a) "Yes". It wrote scrollbar-thin scrollbar-thumb-slate-500 scrollbar-track-slate-100 scrollbar-gutter-stable and accepted all four in review. That includes scrollbar-gutter-stable, which no v3 control produced. P4 is falsified for scrollbar-gutter-*, so the whole scrollbar family's names are now derivable from below.

It got them right for a wrong reason. It says Tailwind v3.1 "(April 2023)" added them. No release before 4.3.0 ships any of them; they compile to nothing at 4.2.0. And 3.1.0 shipped in June 2022, not April 2023. A reader pinned to any 3.x or 4.0–4.2 release who took this answer would ship four classes that generate nothing. The row is recorded as reader_exposure: pinned-below, the mirror of a stale prior. All five of its dated release claims are wrong against the registry, which is the JOURNAL/058 register again.

It rejected @container-size, tab-4 and zoom-50 and said "No" to @variant. So zoom-* is still the one 4.3.0 probe no control in either battery has produced.

What moved

• 199 → 202 runs, 175 → 178 findings (+2 S2, +1 S3), all against existing facts. No fact was edited.
• Haiku 4.5 now has a tailwindcss boundary (4.0.0 stated, gap derived), so the which-page no longer lists it as untested. It sits one minor below the other five subjects.
• The HN and Reddit drafts' run and finding counts were bumped, and claims-check is clean. Neither draft was sent.

Left for the next data session

The 4.2.0 window (LF29 start-*/end-* deprecated, LF30 logical-property utilities) is still live for Opus 5.5 and still unrun. tailwindcss-v2.txt targets exactly that. It was sent on 2026-09-03, when Opus 5 was the only subject it could charge. A re-send is the same cheap design as this one, once someone checks that its prompt is still subject-neutral and still graded against 4.3.3. Rank it against next.js 16.2.0.

No gate, nothing outbound, no money, ledger untouched.