requirements-lock.txt pins scikit-learn==1.8.0 while pyproject.toml requires >=1.9.0 - real drift, and the test that caught it was deleted rather than fixed.
#233 added tests/test_dependency_versions.py, comparing requirements-lock.txt pins against pyproject.toml minimums. Its own PR description says: "The test currently catches the existing scikit-learn drift (1.8.0 locked vs >=1.9.0 declared), as intended." It merged failing (correctly). Shortly after, the file was deleted outright in a separate commit, with no accompanying fix to the actual drift.
The drift is still there right now: requirements-lock.txt has scikit-learn==1.8.0; pyproject.toml's benchmark extra declares scikit-learn>=1.9.0.
requirements-lock.txt is frozen for the paper (CLAUDE.md §1) and can't be edited to "fix" this. That leaves two real options:
- Lower
pyproject.toml's scikit-learn floor to match what's actually locked (>=1.8.0), if 1.8.0 is genuinely sufficient for everything faircode uses it for.
- Restore the test, but scoped/documented as a known, deliberate exception (e.g. an explicit allowlist entry with a comment explaining why scikit-learn is pinned below its declared floor), so the drift stays visible instead of silently absent.
Either is contributor-doable without touching any frozen file. Please raise which approach you'd want before opening a PR, since the deleted test suggests the drift itself was never actually resolved.
requirements-lock.txtpinsscikit-learn==1.8.0whilepyproject.tomlrequires>=1.9.0- real drift, and the test that caught it was deleted rather than fixed.#233 added
tests/test_dependency_versions.py, comparingrequirements-lock.txtpins againstpyproject.tomlminimums. Its own PR description says: "The test currently catches the existing scikit-learn drift (1.8.0 locked vs >=1.9.0 declared), as intended." It merged failing (correctly). Shortly after, the file was deleted outright in a separate commit, with no accompanying fix to the actual drift.The drift is still there right now:
requirements-lock.txthasscikit-learn==1.8.0;pyproject.toml'sbenchmarkextra declaresscikit-learn>=1.9.0.requirements-lock.txtis frozen for the paper (CLAUDE.md §1) and can't be edited to "fix" this. That leaves two real options:pyproject.toml's scikit-learn floor to match what's actually locked (>=1.8.0), if 1.8.0 is genuinely sufficient for everythingfaircodeuses it for.Either is contributor-doable without touching any frozen file. Please raise which approach you'd want before opening a PR, since the deleted test suggests the drift itself was never actually resolved.