R 2.3.0 β†’ Python parity status

This repository freezes the public API against gp3tools R 2.3.0.

Contracts

• Canonical R exports: 278
• Python public namespace: 285 unique names
• Frozen R4 oracle: 3,786 assertions
• Frozen R4 oracle SHA256: 7749F214B3F16A4A03CEBD030C597999B3BDDA38C1BEB9A4CAEDA5380735F684
• Coverage gate: ≥ 90%

Evidence levels

The project deliberately separates API availability from scientific behavioral evidence:

Status	Meaning
EXACT	Frozen deterministic R/Python evidence agrees exactly.
NUMERIC_TOLERANCE	Values agree within a declared numerical tolerance.
STRUCTURAL_EQUIVALENT	Structure and scientific meaning agree while representation differs.
BACKEND_EQUIVALENT	Python uses an appropriate Python backend for the same analysis family.
INTENTIONAL_PYTHON_ADAPTATION	Python-native adaptation is documented and tested.
BLOCKED_OPTIONAL_DEP	Requires an optional backend not installed in the core environment.
FAIL	Measured parity defect.

A whole-surface smoke pass is not presented as universal exact behavioral parity. Exact claims are limited to frozen R oracle evidence.

Behavioral freeze

R1, R2 and R3 are retained as permanent canonical behavioral-parity suites. R4 adds the pupil/QC/workflow/blink/fixation/AOI/geometry family and preserves legacy Python call contracts through semantic dual dispatch. The 3,786-row R4 oracle is frozen and hash-verified; this completion runner does not over-claim a fresh all-row R4 re-score unless a dedicated regenerated scorer is present.

Release gates

The completion runner enforces:

1. 285/278 namespace/export integrity.
2. R1–R3 frozen behavioral regressions plus R4 canonical structure/compatibility gates.
3. Legacy R-alias compatibility.
4. All 278 exports resolving to introspectable callables.
5. Plot export catalog plus headless core plot smoke.
6. Full pytest with ≥90% coverage.
7. Static call-surface audit, Ruff and compileall.
8. Runnable examples.
9. Strict MkDocs build.
10. Wheel + sdist build and isolated wheel import smoke.