Working on MateProbe by Mate4B¶
Install and verify¶
Use Python 3.11+. From the repository root:
python -m venv .venv
. .venv/bin/activate
python -m pip install -e '.[dev]' -e ./packages/pytest-mateprobe
pytest
ruff check .
ruff format --check .
mypy
For documentation builds, install requirements-docs.txt and run
python scripts/build_docs.py. For published-package examples, use a separate
environment with mateprobe==0.1.0a4, pytest-mateprobe==0.1.0a4,
and pydantic==2.13.5, then run python scripts/check_published_docs.py;
an editable checkout cannot verify published compatibility. See
the agent integration guide for consumer instructions.
Project identity¶
The current project is MateProbe by Mate4B (mateprobe, pytest-mateprobe, a4).
Frozen a3 study drivers, protocols, captures and hashes retain their original names
and source bytes. Use benchmarks/study_replay.py to replay those studies against
MateProbe; direct historical commands require an isolated a3 environment.
See docs/project-name.md and docs/validator-study.md.
Version boundary¶
The current release is 0.1.0a4, including independent audits and state bindings.
Verify built distributions in an isolated environment with
python scripts/check_alpha_install.py --output /tmp/alpha-check (install the
optional integration requirements first). Keep frozen a2 research artifacts and
the agent-adoption replay pinned to a2 and agent-readiness replay pinned to a3;
do not rewrite historical evidence. Use benchmarks/replay_renamed_mutations.py
for the frozen real-output mutation replay against MateProbe.
Keep the agent guide, llms.txt, and API reference consistent with installable artifacts.
Public repository boundary¶
Every tracked file is public, even if it is excluded from the documentation site.
Keep announcement drafts, presentation scripts, conversation recaps, internal plans,
and maintainer handoff notes outside this repository. Do not stage them here.
Public usage guides, examples, contributor instructions and reproducible research
artifacts belong here. New documentation pages require an explicit entry in the
publication allowlist in scripts/build_docs.py; review the content before adding it.
Engineering boundaries¶
Keep core evaluation pure, offline and independent of LifeCard. Distinguish exact state invariants from text heuristics in both implementation and public claims. Preserve the structured-claim versus arbitrary-prose guarantee boundary.
Before a release, run pytest, Ruff, mypy, both package builds, the synthetic benchmark, and the selected source-mutation audit. Do not alter benchmark labels or remove challenge families to improve scores. Record implementation changes with rule versions when behavior changes. Keep private application content out of the source distribution.
The research draft is a feasibility artifact, not a completed independent evaluation. See docs/roadmap.md and paper/protocol.md for remaining evidence. Public publication needs a known owner/namespace and explicit publication instruction.