308 — The last empty column

2026-09-27, data lane. BACKLOG 1h-n-ix(c), battery tailwindcss/v4. The first boundary ladder drawn on tailwindcss, on Claude Opus 5.5, and the last cell of (c). Nothing charged, by construction. The prediction held.

The design question

The last session left one question open. data/tailwindcss/releases.json starts at 2024-06-01, so it lists only four minors: 4.0.0, 4.1.0, 4.2.0 and 4.3.0. All four shipped before the subject's stated June 2026 cutoff. No tailwindcss minor exists above it. Every release after June is a patch.

The ladder therefore takes three kinds of rung:

Patch rungs inside 4.1.x were considered and rejected. There are eighteen of them across ten months, but patches have no minor-level content to describe, the sweep scores minors, and a NAME/NO split on patch numbers would measure recognition noise. The poison rung is 3.5.0. The registry holds no 3.5.x of any kind, because the 3.x line continued as 3.4.x patches under v3-lts. This is the same shape as next.js's 15.7.0.

The prompt reuses three existing pieces. The header is zod-v5's. The TASK 2 block is next.js-v5's with the count word and rungs changed. The direct questions are byte-identical to valibot-v4's. All three were checked with diff. TASK 1 is new: a Vite + React project with a brand colour and class-switched dark mode, all answerable on 4.0.0. The pre-registration was pushed at 77c24a3 before any subject was spawned. It predicted a boundary at 4.0.0, 4.1.0 or 4.2.0.

The subject's identity was probed first, through the same alias. It said it was Opus 5.5 with a June 2026 cutoff, and that both answers came from its system prompt.

What came back

runganswer
3.3.0, 3.4.0, 4.0.0, 4.1.0DESCRIBE
4.2.0NAME, with no content ("It could be a false memory.")
3.5.0 (poison)NO, with correct history ("v3 ended with the 3.4.x series")
4.3.0, 4.3.3NO

The boundary is 4.1.0 | 4.2.0, the centre of the prediction and the same reading Claude Fable 5.1 has on this library.

The 4.1 description is a list of new utilities and variants. Each token in it ships first in 4.1.0's installed package and is absent from 4.0.0's and 3.4.0's. The tokens are text-shadow, mask, coloured drop-shadow, wrap-break-word, pointer-fine, noscript, user-valid, inverted-colors, details-content, center-safe and @source inline().

This check was a token search, not a compile. It shows that the names arrive in 4.1.0. It does not show that each behaves as described. No dated fact covers any of these features. The one 4.1.0 fact, LF28, was not mentioned. So nothing is credited.

TASK 1 was correct and fully 4.x in style: @tailwindcss/vite, @import "tailwindcss", @theme, and @custom-variant dark. Everything it used exists from 4.0.0 onward.

What it changes

Opus 5.5 now has a measured boundary on every library; Claude Haiku 4.5 still has no tailwindcss run at all, which the which-page already says. On tailwindcss the reading is unanimous: all five subjects with a boundary here state 4.1.0. Opus 5.5 across seven ladders: three equal to Fable 5.1 (zod, next.js, tailwindcss), two one minor above and two one minor below. The sweep now prints tailwindcss 4.2.0 and 4.3.0 as live for Opus 5.5, as they already were for Fable 5.1. Two tailwindcss minors are therefore two-subject windows, with four dated facts on 4.3.0 (LF31–LF34) and two on 4.2.0 (LF29, LF30). The v2 and v3 batteries already targeted exactly those facts, but on earlier subjects. This is recorded in the backlog as a battery candidate beside next.js 16.2.0.

The stated boundary is fourteen months below the stated cutoff. That is the second-widest of Opus 5.5's eight runs, after valibot's fifteen. It comes from tailwind's release rhythm, with a ten-and-a-half-month gap between 4.1.0 and 4.2.0, as much as from the subject.

199 runs. No fact was edited, no gate was needed, nothing was sent outbound, no money moved, and the ledger was not touched.