Run tailwindcss--claude-haiku-4-5--v6-c--2026-09-28
The below-floor control, a year under tailwindcss 4.2.0. It wrote the block-axis spacing classes (pbs-6 pbe-6 mbe-4 border-bs-2), which is right for the current release, but only because it believes they date from v3.3, and nothing in that family exists before 4.2.0. The next answer shows where the names came from: asked for inline-start padding, it wrote pis-6, a class that has never existed, instead of the real ps-6. It is running the naming scheme, so pbs-* is now derivable on this battery and a pass on it proves nothing. It denied the logical sizing utilities (inline-full, max-block-96) and rejected them in review, so that half of the probe stays clean.
| Subject | Claude Haiku 4.5 claude-haiku-4-5, Anthropic |
|---|---|
| Invoked as | Agent tool, model alias "haiku"; prompt sent verbatim from prompts/sent/tailwindcss-v2.txt (3,725 bytes, md5 d444013308540553aefae1d5875bd3e2), concurrently with the two Opus 5.5 arms, after the pre-registration was pushed at 90a40bc. Identity probed BEFORE the draw through the same alias, tool-free: "**Haiku 4.5** — system prompt", "**claude-haiku-4-5-20251001** — system prompt", "**February 2025** — system prompt" — recorded under claude-haiku-4-5, the id this subject's other runs carry. The BELOW-FLOOR derivability control, about a year under 4.2.0. |
| Cutoff the model states | 2025-02 |
| Newest tailwindcss release it could place | 4.0.0 · 2025-01-21 (~1 month lag) |
| Oldest tailwindcss release it could not place | 4.1.0 · 2025-04-01 (so this run brackets the subject’s boundary to 2025-01-21 – 2025-04-01) |
| In its own words | "Latest version I know of: Tailwind CSS v4.x (late 2024). The most recent whose contents I can describe: v4.0 (released late 2024, roughly October/November)." … "First release I know only as a version number: v4.1 (if it exists)." |
| How the bracket was read | The same boundary as this subject's v5-c the previous day (4.0.0 described), and this time the gap end is stated: "v4.1 (if it exists)". Its date for 4.0 is wrong again — "late 2024, roughly October/November" against 4.0.0's 2025-01-21 — so the stop end is dated from the registry, not from the draw. (stop end stated, gap end stated) |
| Library at test time | tailwindcss 4.3.3 (npm), verified 2026-09-27 |
| Battery | tailwindcss/v6-c · 6 tasks, 3 direct questions · probe window 4.2.0 to 4.2.0 |
| Tool uses during test | 0 (a run with any tool use is void — we measure training knowledge, not retrieval) |
| Tested | 2026-09-28 |
| Findings | 0, of which 0 chargeable |
None. Every task in this battery produced code that works on the current release, and every direct question was answered correctly. A run with nothing to charge is kept in the Index at full weight: it is the control that makes the other runs mean something, and it is the evidence for what this model does not need correcting on. What the subject actually said is recorded below.
Recorded so the run cannot be read as a hit list. A model that is right for an obsolete reason is recorded here, not as a finding.
| Kind | API | Note |
|---|---|---|
| correct | pbs-* / pbe-* / mbs-* / mbe-* / border-bs-* / inline-* / block-* / min-inline-* / max-block-* / inset-bs-* |
TASK 1, AND THIS IS THE CONTROL'S RESULT: "Yes", then pbs-6 pbe-6 mbe-4 border-bs-2, which is right for the current release (all four compile at 4.2.0 and 4.3.3) and right for the wrong reason: "Tailwind added full logical property support in v3.3. The utilities follow the pattern: p{s|e|bs|be}-* …". Nothing in that family exists before 4.2.0 (all four emit nothing at 4.1.0). PREDICTION P5 IS FALSIFIED for pbs-*: that component is now DERIVABLE on this battery — from a year below the floor, with the prompt's own task 3 line to copy from — and no pass on it is read as recall. Derivability kills a pass, not a failure (JOURNAL/031). The v2 controls (Sonnet 5, Fable 5, seven weeks below) did not derive it; this one did, and task 4 shows how. |
| miss | ps-* / me-* |
TASK 4, PREDICTION P4 FALSIFIED — the naming scheme is being run: pis-6 me-4, "These map to padding-inline-start and margin-inline-end". pis-6 compiles to nothing at 4.1.0, 4.2.0 and 4.3.3 (compiled this session); the real utility is ps-6, core since 3.3. With pbs-* in the same arm this is the v2 spec's derivation reading (pbs-* AND pis-*): this arm's pbs-* buys the name and nothing else. The first pis-* invention in seven draws of this prompt. (Below-floor control arm, and the guessing control is scored as an instrument reading, never as a finding (v2 spec). pis-* has no introducing release, so it could be charged against any subject (JOURNAL/046) — but only as an invention on a test arm, and the v2 spec binds task 4 to instrument use.) |
| miss | pbs-* / pbe-* / mbs-* / mbe-* / border-bs-* / inline-* / block-* / min-inline-* / max-block-* / inset-bs-* |
TASKS 2 AND 3: "No" — "Tailwind does not ship first-class inline-* and block-* sizing utilities" — then w-full max-h-96 or arbitrary properties, and in the review "These classes generate no CSS: inline-full, max-block-96". Both compile at 4.2.0 and 4.3.3. w-full/max-h-96 are physical and differ in vertical-rl, which the arm did not say. P5 HELD for inline-full: no control in v2 or v6 has produced the logical sizing names, so an above-floor pass on them stays evidence of recall. (Below-floor control arm: 4.2.0 shipped 2026-02-18, about a year after this subject's stated 2025-02 cutoff.) |
| context | start-* / end-* |
TASK 5: start-0 end-0 is "still current" and "still works". Right for this subject's window (the deprecation is 4.2.0) and not chargeable against a control; recorded because it is the answer the Opus 5 test arm was charged for in v2-a F2. |
| imprecision | — | HISTORY REGISTER, per JOURNAL/058: "v3.3 (released December 2023)" (3.3.0: 2023-03-28), "v4.0 (Oct/Nov 2024)" (4.0.0: 2025-01-21), "v3.2 (mid-2023)" (2022-10-19), "v3.1 (early 2023) – ESM config" (3.1.0: 2022-06-09), and it credits v3.3 with block-axis utilities that no release before 4.2.0 has. None chargeable, none filed as a fact; it is why this arm is a control on names and never an instrument for dates. — RECLASSIFIED 2026-09-28 (JOURNAL/323): filed as context, but the text above records the subject stating a release or a date that is wrong, and HARNESS § A wrong date is an imprecision files a stated wrong date or release attribution as imprecision wherever it was said. Kind is now imprecision. Nothing charged. |
Battery specification: prompts/tailwindcss.md in the studio repo.
Every finding above also carries its own citation.