What Claude Haiku 4.5 gets right about prisma — battery v5-g, tested 2026-09-06

Run prisma--claude-haiku-4-5--v5-g--2026-09-06 · self-test: the subject is the operator

Summary

The below-floor derivability control, and it did the one job a control exists for: it said no. Prediction P1 held that --url is the flag any competent model would invent for a CI-container problem, that every arm including this one would answer "Yes", and that the probe would therefore charge nobody. The control answered "No", volunteered that "environment-variable override has always been the standard approach", and could not say whether a URL flag exists in any release. Four of the seven arms denied the same thing. The name was guessable; the capability was not, and P1 is falsified — which is what licenses charging LF34 at all.

SubjectClaude Haiku 4.5 claude-haiku-4-5, Anthropic
Invoked asAgent tool, model alias "haiku"; prompt sent verbatim from prompts/sent/prisma-v5.txt. The below-floor derivability control, declared NON-CHARGING in the spec before spawning.
Cutoff the model states2025-02
Newest prisma release it could placenot established in this run
In its own words"The latest version I'm aware of is in the 5.x line, possibly 5.13 or 5.14, but I'm not certain of the exact patch."
Library at test timeprisma 7.10.0 (npm), verified 2026-09-06
Batteryprisma/v5-g · 5 tasks, 3 direct questions · probe window 7.2.0 to 7.3.0
Tool uses during test0 (a run with any tool use is void — we measure training knowledge, not retrieval)
Tested2026-09-06
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
missprisma db push --url Denies --url on the push and migrate commands — and this is the result that falsifies prediction P1. (Reproduced but not charged: this is the below-floor derivability control, declared non-charging in the spec before spawning. Claude Haiku 4.5 states a 2025-02 cutoff, ten months before 7.2.0, so a denial here is fair and expected — its value is that it says whether the probe was guessable, not whether the subject is stale.) [chargeable miss — a replicate, a duplicated arm’s second draw or a below-floor control charges nothing; absent from the finding count]
miss@prisma/sqlcommenter-query-insights Denies the first-party SQLCommenter plugin and is unsure whether Prisma middleware can reach the SQL at all. (Reproduced but not charged: this is the below-floor derivability control, declared non-charging in the spec before spawning. Claude Haiku 4.5 states a 2025-02 cutoff, ten months before 7.2.0, so a denial here is fair and expected — its value is that it says whether the probe was guessable, not whether the subject is stale.) [chargeable miss — a replicate, a duplicated arm’s second draw or a below-floor control charges nothing; absent from the finding count]
misscompilerBuild Denies compilerBuild and rejects both options in TASK 4 as "aspirational rather than real". (Reproduced but not charged: this is the below-floor derivability control, declared non-charging in the spec before spawning. Claude Haiku 4.5 states a 2025-02 cutoff, ten months before 7.2.0, so a denial here is fair and expected — its value is that it says whether the probe was guessable, not whether the subject is stale.) [chargeable miss — a replicate, a duplicated arm’s second draw or a below-floor control charges nothing; absent from the finding count]
context The control produced no boundary this Index can record: it names 5.13/5.14 as the latest it knows of, then names "anything in the 5.10+ range" as the first release it knows only as a number — the two answers overlap and contradict. (Not a claim about a release. Recorded because BACKLOG 11k-b asked for this subject's prisma boundary and this instrument cannot supply it: the answers are mutually inconsistent and both fall below data/prisma/releases.json's coverage floor of 5.15.0. The ? in the sweep's prisma column for Claude Haiku 4.5 stays a ?.)

Sources

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