MNE-Python upstream reuse (v0.0.3)

[BabaSanfour]

Track places where mne-denoise could remove duplicated infrastructure only after a corresponding public API/change lands in MNE-Python.

The idea inspire from #91 . With #92 landing a first phase of already public mne functions.

This is from a brainstorming with codex: TBD.

MNE-Python change	mne-denoise target	Change	Impact	Related
Public/generalized GED transformer	dss/linear.py, DSS	Reuse MNE's existing GED infrastructure for generic decomposition, whitening/rank handling, filters/patterns and inverse transform; keep DSS biases/selection local.	Very high	MNE #13392, related PR #14050
Annotations → sample spans/mask	asr/_windowing.py	Public API around _annotations_starts_stops() so MNE owns orig_time, cropping, first_samp, clipping and annotation→sample conversion.	High	New MNE issue needed
Array covariance API	_covariance.py, DSS	Generalize useful parts of _regularized_covariance() if needed by public GED; keep weighted/chunked mne-denoise-specific modes local.	Medium–High	Prefer discussing with GED work
Generic sensor pre-whitening	dss/_whitening.py	Factor reusable pre-whitening from ICA rather than exposing ICA private helpers directly.	Medium	New MNE issue if generally useful
Public/generalized _COLA	_blending.py	Investigate whether MNE overlap-add infrastructure can support mne-denoise's segmented/owned-window semantics.	Medium	New MNE issue if equivalent
Public/generalized _Interp2	asr/_reconstruction.py	Investigate reuse for ASR raised-cosine reconstruction-matrix transitions.	Low–Medium	New MNE issue if exact semantics match

Priority

• Finish SSP-SIR reconstruction upstream work.
• Audit DSS against MNE's GED framework and coordinate with MNE #13392.
• Propose a public annotation → sample-span/mask API.
• Investigate covariance as part of GED.
• Investigate pre-whitening, _COLA, and _Interp2 only if they are generally useful upstream.
• mne.decoding.SpatialFilter for viz

Do not expose private MNE helpers directly just to reduce local LOC. The goal is to establish the right public abstraction in MNE first, then simplify mne-denoise. Not every function needs to be reused.