mne-advanced

MNE Advanced / Rare Methods (grill → analyze → critic)

A hub for the non-standard analyses — microstates, entropy/complexity, graph metrics, the 1/f deep-dive, intracranial specifics, fNIRS GLM, real-time. Each runs mostly via mne_run_code with an external library that is an optional dependency (named per method below). This skill is extra skeptical: a smaller literature means more researcher degrees of freedom and easier over-interpretation, so the discipline is to grill appropriateness and parameter-sensitivity before computing, frame as exploratory, and critique before believing.

Companion skills: mne-mcp-guard for technical execution safety; mne-methodology-critic for Phase 3; mne-spectral for the canonical PSD / aperiodic workflow. Loaded objects persist in one MNE session. Install the named library first (e.g. python -m pip install pycrostates); flag it as an optional dependency to the user before running.

---

Scientific libraries are user-managed. Check imports in the server interpreter first; do not run package installation inside mne_run_code. Environment changes require user authorization.

PHASE 1 — GRILL (before computing anything)

These methods are non-standard, so the intake is harsher. Do not compute until these are answered. If the user can't answer one, propose a sensible default and explicitly flag the open risk — never silently choose.

Appropriateness (the gate question)

• Is this method validated for this data type and question, or would a standard method answer it better? (e.g. "network hubs" via graph theory when a simple ROI contrast suffices; entropy as a vague "complexity" proxy when band power is the real hypothesis.) Justify choosing the rare tool.
• Confirmatory (metric + ROI/band pre-specified) or exploratory (then say so, and frame every number as hypothesis-generating)? With a small literature, default to exploratory.

Parameters & sensitivity (these methods are parameter-hungry)

• What are the key parameters, and how sensitive are results to them? Plan a sensitivity sweep, not a single setting. Examples: microstates n_maps (4–7) + GFP-peak selection; sample entropy m, r, length; multiscale entropy scales; graph threshold / density; HFO detector thresholds; specparam peak_width_limits / aperiodic_mode.