Put scripts/ under black and ruff (#381)

[realmarcin]

Closes #381.

just lint ran black, ruff and mypy over src/ and tests/ only — so every "lint clean" claim in this run's PRs excluded scripts/gtdb_ground.py, the file most of them changed. Reviews kept pointing at it.

What landed

black now formats scripts/ and ruff checks it. 495 auto-fixable findings applied across 46 files. 184 remain, listed by rule in per-file-ignores rather than blanket-suppressed, so new code is held to the full bar and the debt is greppable:

E501 (129)  B007 (14)  S110 (10)  E722 (6)  F841 (4)  F401 (3)  S608 (3)  …

The vendored files nearly went with it

scripts/validate_id_label_correspondence.py and scripts/chem_formula.py are byte-identical across the Mech repos and checked by the vendored-sync gate. black and ruff had rewritten both — 138 insertions in one — which would have broken that gate.

pyproject.toml already excluded the corresponding test files with a comment explaining exactly this, and I read it only after running the formatter. Reverted, both now excluded, and check_vendored_sync.sh confirms all 6 files still match CultureMech@6be694f3.

How the reformatting is verified

scripts/ has no test coverage, so this rests on what is actually available rather than on behaviour:

• all 65 scripts parse;
• the four exercised by tests still produce their expected output (gtdb_ground --name Acetobacter → GTDB:g__Acetobacter, validate_strict → 0 errors, validate_yaml_scalars → 0, validate_gtdb_coherence → 0);
• full suite and every gate green.

That coverage gap is the same one that let #410 — six scripts importing a module that does not exist — go unnoticed.

mypy stays on src/

mypy scripts/ reports 144 errors in 47 files, 44 of them missing third-party stubs. Adding it would mean either mass suppression or an annotation pass over untested code, neither of which belongs in a lint-scope change. Filed with a suggested order as #411.

just validate-all, validate-strict, validate-gtdb-all, validate-scalars, vendored-sync clean; 1226 passed; lint clean over src/ tests/ scripts/.

---

Review round 2 — the CI gate was never widened

The change had zero CI effect. .github/workflows/lint.yaml inlines black --check src/ tests/ rather than calling just lint, so widening the justfile did nothing where it matters. #381 names that file explicitly: the workflow already triggers on scripts/**/*.py and then checked nothing in them. A new script with unsorted imports and 120-char lines would have landed green.

I widened the recipe and never opened the workflow it exists to mirror.

just format was left behind too, so just lint could reject a file the repo's documented fix command would not touch.

One ignore was over-broad. E741 has zero occurrences in linted scripts — its only two live in a vendored file ruff excludes entirely, so the entry could never apply to existing debt and would have silently permitted a rule the rest of the repo enforces. Removed; the comment now states that every remaining code is backed by a real finding. (S110 is 9, not 10 — the tenth is in that same vendored file.)

extend-exclude did not survive a named invocation. black scripts/chem_formula.py reformatted a vendored byte-identical file; only the directory form was protected. force-exclude is set in both tools now — the guard I should have reached for when I first caught this, instead of reverting the file and leaving the mechanism open.

My verification claim was true and empty

All four "test-exercised" scripts are AST-identical or untouched by this diff, so citing them said nothing about the 35 files whose ASTs actually changed. What does cover them, established in review:

• re-running the tools over main reproduces this branch byte-for-byte;
• an identifier census across all 65 scripts differs in exactly 4 expected fix sites, each read and confirmed equivalent;
• no docstring, prompt, report template, regex or CLI help string changed;
• no binary/append open mode changed; every removed import is genuinely unused, and scripts/ contains no side-effect imports;
• no import crossed a sys.path.insert boundary in either direction;
• mypy scripts/ main vs branch shows only removals — zero new type errors;
• all 61 importable scripts produce identical import outcomes on both trees.

Counts corrected

500 fixed (not 495), 180 remaining (not 184), 128 E501 (not 129), 143 mypy errors (not 144 — the delta is a removed import duckdb). These are the reference for #411's cleanup, so they matter.

1226 passed; lint clean over src/ tests/ scripts/ in both the justfile and the workflow.