mne-writeup

MNE Write-up (verified results → manuscript)

The back end of the analysis suite: after the per-category skills have run and mne-methodology-critic has ruled, this skill writes the result up. It mirrors the ARS academic-paper workflow (config → outline → draft → citation/integrity → format) but is specialised for neurophysiology and is gated by the critic.

Companion: mne-methodology-critic (the gate); the per-category analysis skills (the source of results); mne-analyst (the archiving convention this reads from). For a full multi-section paper with literature search and peer review, hand off to the ARS academic-paper / academic-pipeline skills — this skill specialises in the Methods + Results of an MNE study.

⚠️ IRON RULES

1. Only write what PASSED. A result the critic marked BLOCK is not reported as a finding; either fix-and-re-critique first, or move it to Limitations as an acknowledged problem. WARN items must appear as explicit qualifications/limitations, never silently dropped.
2. Every number traces to an artifact. Each reported statistic must come from a file under stats/ (or the analysis tool's output) — never invent or round-guess a value. If it isn't in an artifact, it isn't written.
3. Methods must be reproducible. Pull exact parameters from the equivalent code archived in mne_result/ (filter band, reference, ICA method/components removed, epoch window, rejection threshold, TFR/connectivity/decoding settings, software versions). Vague methods = rejected.

Workflow

For evidence-bound interpretation of current MCP outputs, read evidence-to-claims.md. It adapts the requested academic-paper-strategist -> academic-paper-composer sequence: first align question, method, evidence and permitted claims; then draft and check numbers, terminology and limitations. These external skills are optional for downstream users, not runtime dependencies. Tool-generated interpretation and Results Draft fields are review inputs, never an automatic critic PASS or publication-ready text. Apply the user's preferred language while retaining the exact values, units, direction and inference level from the tool.