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.

1. 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.

1. 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

1. CONFIG (intake)

Confirm: output target (full paper / Methods+Results section / report), discipline + venue, citation format (APA 7 default), language (and whether a bilingual abstract is needed), output format (Markdown → DOCX via Pandoc). Locate the project layout (see references/writeup-guide.md): data/, mne_result/, stats/, critic/, paper/.

2. INTAKE & GATE

Read critic/ verdicts. Build a result inventory: each finding → its figure(s) in mne_result/, its numbers in stats/, and its critic verdict. Drop/flag BLOCKed claims (IRON RULE 1). List exactly what will be reported and what becomes a limitation.

3. OUTLINE

IMRaD (or just Methods + Results). Map every reported result to a figure or table and to its source artifact. Allocate the figures that will be copied into paper/figures/.

4. DRAFT

• Methods — participants/data, recording, preprocessing (filter, reference, montage, bad

channels), artifact correction (ICA method, # components removed and how identified), epoching (window, baseline, rejection + per-group rejection rate), the analysis method and its parameters, the statistics (test, correction, software + versions). Source every parameter from mne_result/.

• Results — report with effect sizes + CIs + corrected p, in APA 7 (see the reporting templates

in references/writeup-guide.md); reference figures/tables; state cluster-level (not point) inference where cluster tests were used.

• Apply a brief writing-quality pass (varied sentence rhythm; avoid AI-typical filler and em-dash

overuse; no throat-clearing openers) — good-writing rules, à la ARS.

5. INTEGRITY & CRITIC GATE

• Citations: every reference cited is in the list and vice-versa; APA 7 formatting; verify DOIs

exist (don't fabricate).

• Methodology gate: re-confirm no BLOCKed claim is reported as a finding, every WARN is

reflected as a limitation, and every number traces to a stats/ artifact. If anything is unverified, dispatch mne-methodology-critic on the draft before proceeding.

6. FORMAT & OUTPUT

Write paper/manuscript.md; convert to paper/manuscript.docx via Pandoc (APA 7); copy/refine the selected figures into paper/figures/; keep references in paper/references.bib. (Reuse the repo's paper/ convention — see MNE_ANALYSIS_SUITE.md for the full project layout.)

See references/writeup-guide.md for the project layout, APA 7 reporting templates (t-test, ANOVA, correlation, cluster-permutation, decoding AUC, connectivity, source), and a reproducible Methods boilerplate.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Exekiel179
• Source: Exekiel179/MNE-MCP
• License: MIT

Install and usage instructions live in the source repository linked above.