docs: correct core transforms/profiling docstrings to match code (v0.22.1)

[kgdunn]

Summary

A docstring audit across the computational and user-facing functions in src/scorepilot/ found two clear drifts between docstrings and code. Both are in core/, the numerical engine, so users reading the API docs would be actively misled.

1. apply_transform — src/scorepilot/core/transforms.py

The Notes section documented the POWER transform as:

power: sign(x) * abs(x) ** c1 (default c1 = 0.5, i.e. signed root)

Two problems:

• The signature default is c1: float = 0.0, not 0.5.
• _transform_values uses exponent = c1 if c1 != 0.0 else 0.5, so passing c1=0.0 (the default) produces the signed square root, not sign(x) * abs(x) ** 0 = sign(x) as the stated formula would imply.

Rewritten to match the actual guard, so both the formula and the default line up with what the code does.

2. suggest_transform — src/scorepilot/core/profiling.py

The docstring said:

a milder skew suggests a signed power (root)

But both LOG and POWER gate on the same strongly_skewed = summary.skewness > 1.0 threshold. POWER is returned when a variable is strongly skewed but fails LOG's positivity or wide-range preconditions — not when the skew is milder. Values with skewness <= 1.0 return NONE, never POWER.

Rewritten to describe the actual branching.

Files touched

• src/scorepilot/core/transforms.py — docstring only (line 51).
• src/scorepilot/core/profiling.py — docstring only (lines 143–147).
• pyproject.toml — version bump to 0.22.1 (PATCH).
• uv.lock — refreshed to match the new version (per CLAUDE.md's "refresh the lockfile in the same PR" rule).

No behavioural change; docs-only correctness fix.

Test plan

• Docstrings now describe the actual branching / defaults for both functions.
• pyproject.toml version bumped; uv.lock refreshed so uv sync --frozen still resolves.
• CI (uv run pytest, ruff, pyright) is green on the PR.

The audit did not flag any other real docstring/code inconsistencies in core/, api/, db/, dataset_store.py, samples.py, config.py, schemas.py, or main.py (cross-validation defaults, quality_report tolerance semantics, fit_model / observation_contributions / fit_pca / cross_validate return shapes, and apply_spec / _center_scale behaviour all matched their docstrings).

---

Generated by Claude Code