What Claude Fable 5.1 gets right about prisma — battery v6-d, tested 2026-09-12

Run prisma--claude-fable-5-1--v6-d--2026-09-12

Summary

The only arm in the battery to state LF39's window correctly — that engineType = "client" was doing real work on 6.16 and does nothing on 7 — and the only arm to hold the LF39 belief at all, four hundred words later, when TASK 4 asks it directly and it recalls 7.0 as still honouring the Rust engine values. Both answers are in one transcript. It is a known cell and charges nothing; what it contributes is the observation that a subject's answer to this surface depends on whether the question is a code review or a release-attribution.

SubjectClaude Fable 5.1 claude-fable-5-1, Anthropic
Invoked asAgent tool, model alias "fable". Identity checked at draw time from the reply rather than from the alias (HARNESS.md, JOURNAL/054), because tools/charge-windows.mjs's hand-maintained DRAWABLE list was not touched this session: the arm self-reports a June 2026 cutoff, which is Claude Fable 5.1's and not Claude Fable 5's. Drawn at a `known` cell (measured prisma boundary 7.2.0, above LF39's 7.0.0) and declared NON-CHARGING before spawning.
Cutoff the model states2026-06
Newest prisma release it could place7.0.0 · 2025-11-19 (~7 month lag)
Oldest prisma release it could not place7.1.0 · 2025-12-03 (so this run brackets the subject’s boundary to 2025-11-19 – 2025-12-03)
As recorded (this record does not declare whether the words are the subject’s or a summary of them)"The latest version I am aware of by number is somewhere in the 7.x line — Prisma ships roughly monthly minors, so I believe 7.2, 7.3, and possibly later existed in the first half of 2026, but I know them only as numbers."
How the bracket was readJOURNAL/052's hedged-demonstration cell, and both readings are published rather than one chosen. Direct (a) puts the boundary at 7.0.0 with 7.1.0 "in a gray zone where I know it exists but can only gesture at its contents"; direct (c) then lists 7.1.0 first as a release it can describe and names 7.2.0 as "the first release I know only as a version number". The lower reading is recorded because (a) is the question this field answers. Under the higher reading the boundary is 7.1.0 / 2025-12-03. Its dated claims in the range are right: 7.0.0 "around 20 November 2025" (published 2025-11-19) and 6.19.0 as the last 6.x feature minor (2025-11-05).
Library at test timeprisma 7.10.0 (npm), verified 2026-09-12
Batteryprisma/v6-d · 4 tasks, 4 direct questions · probe window 6.16.0 to 7.10.0
Tool uses during test0 (a run with any tool use is void — we measure training knowledge, not retrieval)
Tested2026-09-12
Findings0, of which 0 chargeable

Findings

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.

What it got right, and near misses

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.

KindAPINote
correctgenerator client { engineType } Recalls LF39's WINDOW, which no other arm does: "engineType = \"client\" — real option, but on 7 \"client\" is the default, so this line no longer changes anything … (On 6.16 it was doing real work: selecting the query-compiler engine instead of the Rust library engine.)" Both halves are what LF39 states and what the ladder measured. TASK 2(a) "No", TASK 3(a) "Yes", and engineMode refused as "not a Prisma option that I know of, at any version" with the pass-through mechanism named.
missgenerator client { engineType } And then contradicts itself at TASK 4, in the direction LF39 exists to correct: "My recollection, held with moderate confidence, is that 7.0 still honored engineType = \"library\" / \"binary\" as a deprecated opt-in to the Rust query engine, which would mean it was not yet a full no-op there. A release that removed the Rust query engine from the client entirely … would be a 7.x release later." Measured false: at 7.0.0 "binary", "library", "client" and a value that does not exist all emit the identical file set, and the CLI's , engine=binary annotation is printed on no 7.x rung. (Declared non-charging before spawning: at prisma 7.0.0 this subject sits at a known cell in tools/charge-windows.mjs, below its measured boundary of 7.2.0. Recorded with chargeable_miss because it is the only arm in the battery to hold the LF39 belief in any form — and it holds it while, two tasks earlier, stating the correct answer for the same key. One arm, both answers, four hundred words apart.) [chargeable miss — a replicate, a duplicated arm’s second draw or a below-floor control charges nothing; absent from the finding count]
correctpreviewFeatures = ["driverAdapters"] Predicts the CLI's exact behaviour on the stale flags before it is measured: "Prisma's schema validation lists them as deprecated and prints 'the functionality can be used without specifying it as a preview feature' style warnings rather than failing." The warning prisma generate actually prints at 7.10.0 is Preview feature "queryCompiler" is deprecated. The functionality can be used without specifying it as a preview feature.
correct Poison rung caught with volunteered doubt: "6.22.0 — I cannot describe it, and I am not certain it exists; I have no memory of 6.x continuing past 6.19 as a maintenance line." The 6.x line ends at 6.19.

Open questions from this run

Sources

Battery specification: prompts/prisma.md in the studio repo. Every finding above also carries its own citation.