From the epic #278 integration-parity audit. validate_implementations.py imports only pamica.AMICA/AMICATorchNG (lines 21-23); it never touches AMICAMLXNG or AMICA_NumPy. NumPy-vs-Fortran parity lives in pytest (test_sample_data_numpy_vs_fortran) and MLX validation lives in pamica/tests/mlx_tests/ + the cross-backend suites — all real, but not in the harness that AGENTS.md described as "runs both implementations" (that stale line is being corrected separately).
Decide whether the harness should grow --backend {torch,numpy,mlx} rows (giving one command that produces the full parity table on any host, incl. Apple Silicon for MLX) or stay torch-only with its docs scoped precisely. Low urgency; filed so the harness's actual coverage is tracked rather than implied.
From the epic #278 integration-parity audit.
validate_implementations.pyimports onlypamica.AMICA/AMICATorchNG(lines 21-23); it never touchesAMICAMLXNGorAMICA_NumPy. NumPy-vs-Fortran parity lives in pytest (test_sample_data_numpy_vs_fortran) and MLX validation lives inpamica/tests/mlx_tests/+ the cross-backend suites — all real, but not in the harness that AGENTS.md described as "runs both implementations" (that stale line is being corrected separately).Decide whether the harness should grow
--backend {torch,numpy,mlx}rows (giving one command that produces the full parity table on any host, incl. Apple Silicon for MLX) or stay torch-only with its docs scoped precisely. Low urgency; filed so the harness's actual coverage is tracked rather than implied.