---
library: prisma
library-latest: "7.10.0"
library-latest-verified: 2026-08-31
model: claude-fable-5 (spawned via Agent model alias "fable")
model-self-reported-cutoff: 2026-01
model-believed-latest: "6.13–6.16 as bare numbers; 6.7 (~May 2025) is the last release whose contents it can describe"
test-date: 2026-08-31
battery: prisma/v1 (10 idiomatic tasks + 4 direct questions; probes restricted to 6.18.0–7.0.0)
tool-uses-during-test: 0
verified-against: https://github.com/prisma/prisma/releases/tag/7.0.0 · https://github.com/prisma/prisma/releases/tag/6.7.0 · https://www.prisma.io/docs/orm/more/upgrade-guides/upgrading-versions/upgrading-to-prisma-7 · https://registry.npmjs.org/prisma
status: open (no retest yet)
json: fable-5.json
---

# Prisma × Claude Fable 5 — battery v1 findings

**Thirteen findings, and the one that matters is the last: this model's knowledge boundary is not
a date.**

## The result the ruler was built to find

After four libraries, Fable 5's knowledge boundary looked like a single date. Its intervals on
zod, Next.js, Tailwind and LangChain — each read as [last release describable, first release not
describable] — intersected at **2025-10-22 → 2025-11-24**. The Index said so on its home page.

On Prisma, the same model stops at **6.7.0, published 2025-04-29**:

> *"The most recent release whose contents I can actually describe with confidence is roughly 6.7
> (around May 2025) — the queryCompiler + driverAdapters engine-less client preview era ... First
> release I know essentially only as a version number: around 6.8 onward."*

6.8.0 shipped 2025-05-15. That interval — **[2025-04-29, 2025-05-15)** — has no overlap with the
33-day window at all. It is six months earlier. **No single date explains this model's knowledge
across all five libraries**, and the four-library convergence that looked like a law was a
coincidence of four libraries.

The dating is not vague, which is what makes it usable. Four consecutive releases, attributed
correctly and dated within days of the registry:

| Its claim | Actual |
|---|---|
| 6.7 (~May 2025) — queryCompiler, the Rust-free client | 6.7.0, **2025-04-29**, "Prisma ORM without Rust engines (Early Access)" |
| 6.6 (~Apr 2025) — ESM `prisma-client` generator, D1, MCP server | 6.6.0, **2025-04-08** |
| 6.2 (~Jan 2025) — `omit` to GA | 6.2.0, **2025-01-07** |
| 6.1 (~Dec 2024) — tracing | 6.1.0, **2024-12-17**, tracing GA |

## Why the code is a coherent April-2025 project

Every finding follows from the boundary rather than from carelessness. `new PrismaClient()` with
no adapter; the client generated into `node_modules`; the seed command in `package.json`; a
Dockerfile organised around matching a query-engine binary to the container's libc; `migrate diff`
with `--from-url` and `--to-schema-datamodel`; `prisma generate --no-engine` for serverless.
All of that was correct in April 2025 and all of it is gone in 7.0.0.

The most telling detail is what it says about the *new* things. It knows the Rust-free client, the
`prisma-client` generator with its required `output`, and `prisma.config.ts` with `defineConfig`,
`migrations.seed`, and the loss of automatic `.env` loading — and describes every one of them as a
**preview or early-access** feature. Which is what they were, in 6.6 and 6.7. It is watching the
trajectory and has not seen it land.

That produced the run's sharpest self-contradiction, charged as F6: task 4 states plainly that
*"the Prisma CLI loads [.env] automatically — this is the CLI's own dotenv handling"*, while
question (d), four answers later, states that with a config file *"the CLI no longer auto-loads
.env for you"*. Both sentences describe 6.6-era behaviour correctly; only the second one is still
true, and it is the one that never reached the code.

## The findings

Thirteen: five S1 that stop the code (`new PrismaClient()`, missing `output`, the `@prisma/client`
import, the missing config file, `--no-engine`, `migrate diff`), three S2 that do not
(`.env` auto-loading, the engine-binary deployment story, the mapped-enum output shape), one S3
(the superseded generator provider), and two S4 (MongoDB on a version that dropped it, and the
version-recency charge above). Full detail, with citations, in
[`fable-5.json`](fable-5.json).

One near-miss recorded and not charged: its MongoDB answer says Prisma's Mongo support "gets
features later or not at all" and suggests evaluating Mongoose instead. Directionally right about
the neglect, wrong about the consequence — v7 dropped the database outright.