mne-preprocess

MNE Preprocessing (grill → analyze → critic)

Preprocessing & data-quality cleanup of neurophysiology data via the MNE MCP server. This skill is skeptical by design: nearly every preprocessing mistake (a too-aggressive high-pass that eats your slow ERP, a reference that biases connectivity, filtering after epoching, differential bad-channel handling between groups) runs without any error and silently biases everything downstream — so the discipline is to grill the downstream analysis before cleaning, and critique the pipeline before believing.

Companion skills: mne-mcp-guard for technical execution safety; mne-methodology-critic for Phase 3. Loaded objects persist in one MNE session, and preprocessing edits are in place.

---

PHASE 1 — GRILL (before processing anything)

Do not filter, reference, or resample until these are answered. If the user can't answer one, propose a sensible default and explicitly flag the open risk — never silently choose.

The question that decides every parameter

• What is the DOWNSTREAM analysis? Filter choices follow from it, and there is no neutral default: 0.1 Hz high-pass for ERP (slow components survive), ~1 Hz high-pass for ICA (drift hurts decomposition), broadband / no/low high-pass for TFR / low-frequency power. Set the pipeline to the analysis, not the other way round.

Filtering

• High-pass edge — and ⚠️ will the filter edge distort the effect of interest? A 0.5–1 Hz HP can attenuate/shift slow ERP components (CNV, P300, readiness potential). Justify the edge.
• Low-pass edge (anti-alias / smoothing) and transition bandwidth; FIR (linear-phase, default) vs IIR (causal, ringing tradeoffs)?
• Line frequency: 50 or 60 Hz? (region-dependent) — notch it (+ harmonics) or rely on a low-pass below it?

Resampling

• Needed at all? If so, before or after epoching? Downsample after epoching where possible — resampling continuous data jitters event sample positions. New sfreq must stay > 2× the low-pass (anti-aliasing) — MNE low-passes on resample, but verify.