fix(derivations): evaluate() returns the no-spread result for every bin method on a constant column

[cnicholas]

Summary

• What: the no-spread guard in _evaluate_bin now also treats non-strictly-increasing fitted edges as degenerate, so all three automatic bin methods return the documented no-spread result on a constant column.
• Why: equal_freq already returned it (np.unique collapses the quantiles to one edge). equal_width fitted identical finite edges and sd fitted a zero-sigma edge set; both passed the guard and pd.cut raised ValueError: Bin edges must be unique. The module docstring promises evaluate never raises on a routine state, and a constant column is one. Found through the app on a single-time-point file whose TIME column was all 1.
• Scope: one guard in derivations.py, the no-spread fitted dict shape, three tests, CHANGELOG.

Contract / invariants

• Defaults unchanged — non-degenerate fits are byte-identical: edges, labels, and the nextafter top-edge widening are untouched. validation/e2e_bishop_report.py exits 0.
• Residuals unaffected — derivations only.
• Row/index alignment preserved — the no-spread result already had the input's index; unchanged.
• Output schema compatible — the no-spread fitted dict gains n_bins: 0, edges: [], labels: [] (additive; validate() reads n_bins with .get).
• Pinned error strings unchanged — 'column has no spread; cannot bin' is unchanged.

Behaviour changes

• evaluate() returns instead of raising for equal_width / sd on a constant column.
• The no-spread fitted dict carries a consistent shape. Explicit breaks on a constant column still bin (user cut points are strictly increasing; everything lands in one interval), covered by a test.

Methodology

• No methodology change — implementation only.

Tests

• test_constant_column_is_no_spread_for_every_method (parametrized over equal_freq, equal_width, sd) — no raise, no-spread message, zero fitted bins, all-NA categorical with no categories.

• test_constant_column_with_explicit_breaks_still_bins — breaks are unaffected.

• test_constant_column_validates_without_raising — validate() calls evaluate() internally and must survive too.

• pytest tests/ — 2340 passed, 10 skipped

• ruff check . — clean

• mypy processbehavior/derivations.py — clean

• Golden masters untouched

Notes

The companion app fix (the preview panel must not take the page down on empty counts) is separate and works against the 0.2.0 pin; this change makes the library side honest regardless of caller.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BKphCi1LC1hjGYteWEK9JV