chore: pin lint tooling + normalize source and hand-fixed test files (1/2)

[igerber]

Summary

• Part 1 of 2 of the repo-wide lint normalization, split so each PR fits the AI reviewer's input cap (the combined ~1.18MB diff exceeded the 1MB Codex input limit and crashed the review run; the mechanical-only remainder is PR 2/2). Together the two PRs reproduce the original change byte-for-byte.
• Pin the lint toolchain exactly in the dev extra (black==26.3.1, ruff==0.15.13, mypy==2.1.0) so every machine and the upcoming Lint CI workflow judge cleanliness identically. The tools require Python >= 3.10 (dev tooling only; the library floor stays 3.9).
• Scoped [tool.ruff.lint.per-file-ignores] for deliberate patterns, each with a why-comment: trop* logger-before-imports (E402), honest_did Rambachan-Roth math notation l (E741), __init__ backend re-export surface (F401), conftest sys.path ordering (E402), tutorial-quoted CdH horizon notation (E741).
• Mechanical normalization of diff_diff/ source (commit 1): verbatim ruff check diff_diff --fix + black diff_diff output at the pinned versions.
• The 16 test files with non-autofixable findings (commit 2): their mechanical reformat plus hand fixes - 2 latent F821s (typing names used in an annotation without import), 1 import hoist (E402), 1 rename (E741), and 26 individually audited F841 unused locals, two of which were real never-asserted test bugs now upgraded to behavioral assertions (test_trends_nonparam_basic asserts r_plain finiteness; test_influence_function_normalization drops an abandoned R-comparison block and gets an honest docstring).

Reviewer note: commit 1's diff_diff/ changes are byte-reproducible with the two commands above; all judgment calls are in commit 2.

Methodology references (required if estimator / math changes)

• Method name(s): N/A - no methodology changes (formatting/lint only; mypy error count verified byte-identical at 184 pre/post, confirming no semantic drift in diff_diff/)
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: no new tests; 16 test files hand-fixed. Targeted runs green across all touched areas, including the full tests/test_openai_review.py -m '' (250 passed).
• Backtest / simulation / notebook evidence (if applicable): N/A

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

🤖 Generated with Claude Code

https://claude.ai/code/session_01Lbd6nqWmg4snvvBmegwqiw

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 6d77942b907965d21142847572a05b6494a3ee8c

---

Overall Assessment

✅ Looks good — no unmitigated P0 or P1 findings.

Executive Summary

• The changed library source appears limited to import ordering, formatting, and unused-import cleanup.
• I found no estimator math, weighting, variance/SE, identification assumption, or default behavior changes.
• Methodology registry/TODO conventions were checked; no new undocumented methodology deviation is introduced.
• diff_diff source files touched by the PR parse cleanly with ast.parse.
• Runtime import/tests could not be run in this sandbox because numpy is not installed.

Methodology

No findings.

Affected methods: none. The estimator-touching files in the supplied diff do not alter formulas, SE paths, inference gates, weights, assumptions, or defaults. The HonestDiD reference remains unchanged in diff_diff/honest_did.py:L1-L17, and the registry remains the controlling methodology source at docs/methodology/REGISTRY.md:L1-L5.

Code Quality

No findings.

The suspected import/syntax issue in diff_diff/__init__.py is not present; the initializer is structurally valid around the reordered exports at diff_diff/__init__.py:L23-L303. The removed local warnings import in TROP is safe because the module-level import remains at diff_diff/trop_local.py:L13-L15.

Performance

No findings.

The source changes do not introduce new loops, allocations, estimator execution paths, or backend dispatch changes beyond mechanically split imports.

Maintainability

• Severity: P3 informational
  Impact: The dev extra now pins tools that are documented as requiring Python >= 3.10 while the package runtime still targets Python 3.9. This is already called out in both config and changelog, so it is not a blocker.
  Concrete fix: No required change. Optional: split lint/type tooling into a lint extra or contributor-doc note if Python 3.9 contributors need .[dev] to install cleanly.
  Location: pyproject.toml:L63-L69, CHANGELOG.md:L11-L17

Tech Debt

No findings.

No new deferred methodology, correctness, or test debt is introduced by the reviewed diff. Existing deferred items remain tracked under TODO.md:L50-L80.

Security

No findings.

The supplied diff does not add secrets, credential handling, network behavior, or code execution surfaces.

Documentation/Tests

• Severity: P3 informational
  Impact: I could not independently run the test suite or import smoke test in this sandbox because numpy is unavailable. I did verify the touched source files parse cleanly in-memory.
  Concrete fix: Ensure CI runs the touched test set in a dependency-complete environment. No PR code change required.