Audit and stabilise the mathematical core

[DiogoRibeiro7]

Audits every mathematical quantity the package exports before new features are layered on top.

Added

tests/test_mathematical_core.py — 41 hand-computable tests covering membership representation, the fuzzy operators, sufficiency, necessity, subset relations, calibration, truth-table classification and minimisation. Each assertion carries the arithmetic in a comment so it can be checked against the definition without running the code.

Boundary and degenerate behaviour is now pinned explicitly:

Situation	Behaviour
Σ X = 0	consistency, coverage and PRI are 0, not NaN
Σ Y = 0	coverage and necessity consistency are 0
Σ (1 − min(X,Y)) = 0	RoN is 0
membership exactly 0 or 1	admissible; operators reduce to crisp logic
NaN or infinite	ValueError — never imputed
outside [0, 1]	ValueError — never clipped

docs/mathematical_validation.md — the audit table: object, equation, implementation, test, tolerance and validation status, plus full definitions for the calibration transformations and row coding.

Findings

1. Unique coverage is not implemented. Raw coverage is; the per-term unique coverage other QCA software reports is absent. Recorded as a gap, not a divergence.
2. Intermediate solutions are not the standard algorithm and stay marked experimental.
3. No untested mathematical helper remains in the metric or minimisation layers.
4. Returning 0 for an undefined ratio is documented as a deliberate convention — propagating NaN would silently poison downstream aggregation.

No behaviour changed; this is verification and documentation only. 177 tests pass at 100% coverage.

---

Summary by cubic

Audit the mathematical core of setqca with hand-computable tests and clear boundary conventions to stabilize metrics, operators, calibration, truth tables, and minimisation. No behavior changes.

• New Features
  • Added tests/test_mathematical_core.py with 41 hand-computable tests covering membership, fuzzy operators, subset relations, sufficiency/necessity (consistency, coverage, PRI, RoN), calibration, truth-table coding, and minimisation.
  • Added docs/mathematical_validation.md with equations, implementation links, test references, tolerances, and validation status; linked in mkdocs.yml under Validation.
  • Pinned boundary behavior: empty cause/outcome return 0 (no NaNs), RoN=0 when denominator is 0, crisp 0/1 allowed, values outside [0,1] or NaN/∞ raise ValueError (no clipping or imputation).

^{Written for commit 77bb834. Summary will update on new commits.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!