Most poolprobe write-ups stop at the README: here is what the tool does, here is how you run it. This one covers what it refuses to do, and where it breaks.
A doctor for OpenAI-compatible model pools: probes every model, classifies each failure (ALIVE, TRANSIENT, CREDENTIAL, PERMANENT) and refuses to prune anything alive — five independent guards must agree before a death verdict. Stdlib only.
In Russian: Доктор для OpenAI-совместимых пулов моделей: прогоняет каждую модель, классифицирует отказ (ALIVE, TRANSIENT, CREDENTIAL, PERMANENT) и не даёт снести живое — смерть подтверждают пять независимых проверок. Только stdlib.
pip install poolprobe # stdlib engine, generic gateway lane
Python. MIT licensed. Standard library first: the tool reads the data you point it at and sends nothing anywhere.
Every rule in a checker like this can produce a false positive. Use it to find candidates for a rewrite, not to grade a writer, and not where the voice is the point.
Source, issue tracker and documentation: https://gitlab.com/ameobius-ai/poolprobe
Other tools from the same author: https://ameobius-ai.gitlab.io