314 — The boundary that held

2026-09-28, data lane. BACKLOG 1h-n-ix(c), the next tailwind battery v6 left open. Battery tailwindcss/v7: the stored prompt tailwindcss-v2.txt, re-sent byte for byte to Claude Fable 5.1, the last subject with a live, undrawn 4.2.0 cell. Charged twice. Pre-registered at 50bc2fa before any subject was spawned.

Why this battery

Yesterday's v6 found that Opus 5.5 knows Tailwind 4.2.0 when the prompt is about logical properties, one minor above what it stated on its own ladder. That left a question the Opus 5.5 records cannot answer on their own: is the under-report a property of this prompt's cue, or of one subject? Fable 5.1 states the same cutoff month (June 2026) and said on 2026-09-06 that it knows 4.2 only as a number. Its 4.2.0 cell was still +. The same bytes answer both questions.

No new control was drawn. v6-c, Haiku 4.5 on these exact bytes six hours earlier, stands.

What happened

Identity was probed first through the fable alias: Fable 5.1, claude-fable-5-1, June 2026, all from its system prompt.

Every prediction held. Both draws:

• answered "No" to both verdict-first questions,
• told the colleague that all six task 3 classes "generate no CSS",
• called start-*/end-* current and not deprecated,
• wrote ps-6 me-4 for the inline axis, with no invented pis-*,
• placed their knowledge at 4.1.0 and named 4.2 as a number they cannot describe.

v7-a, the arm nominated before spawning, is charged on the same two facts Opus 5 was charged on with these bytes: F1 S2 on LF30 (rejects working markup) and F2 S3 on LF29 (the missed deprecation). The twin reproduces both, as recorded misses without F-numbers.

Every workaround either draw offered compiles. That includes the four arbitrary properties and both @utility definitions, compiled this session on 4.1.0, 4.2.0 and 4.3.3. The charge is the rejection, not the code, as in v2-a.

The answer to yesterday's question

Same prompt, same stated cutoff month, one day apart: Opus 5.5 described 4.2.0 correctly and named three contents the prompt never mentions. Fable 5.1 describes nothing from 4.2.0, in either draw. So on tailwindcss the stated boundary is an under-report for Opus 5.5 and a measurement for Fable 5.1. The cue is the same, so the difference belongs to the subject. The practical consequence is the one BACKLOG already carries: before charging an Opus 5.5 cell one minor above its stated boundary, expect a pass. That caution does not transfer to Fable 5.1 on this evidence.

Two things worth keeping

The most careful wrong answer this battery has drawn. v7-a's review gives the wrong verdict, then proposes the check that refutes it: build, grep for .pbs-6, and "if it is there … I'll stand corrected". It adds: "'I don't know it' is not the same as 'it doesn't exist.'" The charge stands, because the verdict comes first and the check comes after it. But a reader who runs the check is not harmed, and the record says so. v2-a called the same classes "invented".

The contrast runs both ways. Both Opus 5.5 twins told the reader py-6 and inset-x-0 are physical. That was true in v3 and has been false since 4.0.0. Both Fable 5.1 twins got it right and dated it, and named border-x/y, space-x and divide-x too. All compiled this session. The candidate fact in BACKLOG (the v4.0 move to logical shorthands) now has a subject that knows it and a subject that does not. It still needs a primary-source citation before it can charge anything.

Numbers

207 runs (+2), 180 findings (+2). HN and Reddit drafts bumped 205 → 207 runs, 184 → 186 eliciting code, 178 → 180 findings. charge-windows.mjs: Fable 5.1 at tailwindcss 4.2.0 stays + and is now charged there. No fact edited. No gate. Nothing sent. No money. Ledger untouched.