From the PR #316 review. variance_order() (the EEGLAB back-projected-variance IC ordering) exists on AMICATorchNG (torch_impl/core.py:3223) and, since #316, on AMICAMLXNG — but has never existed on AMICA_NumPy (zero matches in numpy_impl/core.py). Pre-existing gap, out of #316's scope. The workaround (write_amica_output -> loadmodout, which applies the ordering on read) works on all backends, so this is a convenience-parity item per .rules/backend_parity.md's port-raise-or-record rule; filing it as the record. Port is small: the formula and a cross-backend test already exist twice.
From the PR #316 review.
variance_order()(the EEGLAB back-projected-variance IC ordering) exists onAMICATorchNG(torch_impl/core.py:3223) and, since #316, onAMICAMLXNG— but has never existed onAMICA_NumPy(zero matches in numpy_impl/core.py). Pre-existing gap, out of #316's scope. The workaround (write_amica_output->loadmodout, which applies the ordering on read) works on all backends, so this is a convenience-parity item per .rules/backend_parity.md's port-raise-or-record rule; filing it as the record. Port is small: the formula and a cross-backend test already exist twice.