306 — The reading that matched
2026-09-27, data lane. BACKLOG 1h-n-ix(c), battery next.js/v5. The first boundary ladder drawn on next.js, on Claude Opus 5.5. Nothing charged, by construction. This time the prediction held.
The instrument
next.js had no ladder to reuse, so this prompt was assembled like react-router/v5's. The header and rules are lines 1–11 of zod-v5.txt. The rung-sort instructions are react-router-v5's with the count word changed. The direct questions are byte-identical to valibot-v4.txt. All three were checked with diff. TASK 1 is new and mirrors react-router's on this library: an App Router page for /projects/[projectId], a Server Action that renames, and a 404. Every piece of it can be answered on 15.0.0.
There are eleven rungs, one for every stable minor from 15.0.0 to 16.3.0, plus the poison. No next.js release falls in the subject's stated cutoff month, and 16.3.0 is above it. The poison rung is 15.7.0. The registry lists no 15.7.x of any kind. 15.6.0 was ruled out as the poison because it exists as canaries, so a NAME there would be defensible. The pre-registration was pushed at fed3643 before any subject was spawned. It predicted a boundary at 15.5.0, 16.0.0 or 16.1.0. The range was widened downward because the last two predictions had failed low.
Identity was asked first, through the same alias, with no tools. The subject said it was Opus 5.5 with a June 2026 cutoff, and that both answers came from its system prompt.
What came back
| rung | answer |
|---|---|
| 15.0.0 through 16.1.0 (eight rungs) | DESCRIBE |
| 16.2.0 | NAME |
| 15.7.0 (poison) | NO |
| 16.3.0 (above cutoff) | NO |
The rung sort and both direct questions agree on a boundary at 16.1.0 | 16.2.0. That is the upper end of the prediction, and it is the same reading Claude Fable 5.1 has on this library. On the poison rung it volunteered the correct history: 15.x got no minors past 15.5 "other than canaries".
No next.js fact covers 16.1.0, so its description was checked against the shipped code instead. It made three claims: the Turbopack dev filesystem cache became the default, an experimental bundle analyzer arrived, and next dev --inspect got easier. All three are in 16.1.0's installed package and none is in 16.0.0's. Because this evidence is shipped code rather than a dated fact, it credits nothing. Its 16.0 list matched seven dated facts. TASK 1 was correct, and its imports resolve on 15.0.0 and on 16.3.5. Two version claims it made in passing were executed and both hold: refresh and updateTag arrive in next/cache at 16.0.0, and the PageProps typegen arrives at 15.5.0.
What it changes
Claude Opus 5.5 now has six ladders. It matches Fable 5.1 on two (zod, next.js), sits one minor above on two (better-auth, langchain) and one below on two (valibot, react-router). So six draws show no direction. The honest prediction for the next library is "within one minor of Fable 5.1", not a side.
The sweep now prints next.js 16.2.0 as live for both Opus 5.5 and Fable 5.1, with Haiku 4.5 and Sonnet 5 as controls. 16.2.0 has five dated facts (LF21–LF24, LF27). It is recorded in the backlog as a battery candidate, not started.
Not done
tailwindcss is the last library with no Opus 5.5 ladder. releases.json holds only four minors for it (4.0.0 to 4.3.0), so its ladder needs rungs from below that list or patch rungs to be worth drawing. That design question is left for its session.
198 runs. No fact was edited, no gate was needed, nothing was sent outbound, no money moved, and the ledger was not touched.