[Bug]: DSR skew/kurtosis use mixed moments, rolling attribution betas shift after NaN drop, Hill lower tail includes gains, Polars 2 timestamp and empty-frame crashes #89

Affected package version

0.1.8 (main at 2a1c3490)

Environment

Linux, source checkout of main at 2a1c3490 with the repo's dev environment (uv sync); offline test suite.

Observed behavior

1. Return skewness/kurtosis divide 1/n moments by the n−1 std, understating deflated-Sharpe inputs on short samples.
2. Rolling attribution fits betas on uncleaned factor rows after NaN return rows are dropped, shifting lagged betas.
3. The lower Hill tail uses absolute returns, so gains enter the loss tail; with that fixed, tail="both" raises on all-positive samples.
4. String factor timestamps fail under Polars 2 (cast rejected); empty frames crash outlier filtering (Null-dtype abs()).
5. Non-mapping YAML/JSON configs raise TypeError instead of the documented ValueError.

Expected behavior

1. Population moments, exact match with scipy.stats.skew/kurtosis(fisher=False). 2. Betas estimated on the same cleaned rows. 3. Loss-only lower tail; both falls back to the upper tail. 4. ISO strings parse; empty frames pass through. 5. ValueError.

Minimal reproduction

Each case has a focused unit test on quantum-like/diagnostic branch review/fixes (one commit per fix, on top of 2a1c3490). Every test fails on main and passes with its fix:

• tightened tests/test_evaluation/test_moments.py::test_reference_scipy_skewness and ::test_reference_scipy_kurtosis to an exact SciPy match on 30 observations
• tests/test_evaluation/test_factor_attribution.py::test_nan_row_matches_predropped_sample
• tests/test_evaluation/test_distribution.py::test_hill_lower_tail_ignores_right_tail_spikes
• tests/test_evaluation/test_distribution.py::test_hill_both_on_positive_sample_uses_upper_tail
• existing TestUtf8Timestamps (both tests were failing before)
• existing test_empty_dataframe (was failing before)
• tests/test_config/test_io.py::test_non_mapping_documents_raise_value_error

Compatibility impact

Public behavior or results

Acceptance criteria

Each case has a test that fails on main and passes with the fix (new tests, tightened SciPy reference tests, or previously failing existing tests).

I have a PR ready with these fixes and will link it here.

Amir