eeg-audit

What problem does it solve?

It prevents incorrect EEG claims from slipping into a report by independently re-checking plan traceability, statistical reporting, figure integrity, and COBIDAS-MEEG method completeness.

Core Features & Use Cases

• Cross-file numeric consistency auditing: Confirms p-values, effect sizes, permutation counts, seeds, and tails match across the analysis plan, stats JSON outputs, and the report text.
• Integrity guardrails via deterministic checks: Runs an mtime verification to detect post-hoc edits of the frozen analysis plan relative to stats generation.
• External “hostile” methods review: Hands the bundled evidence to an external LLM reviewer (Codex MCP, GPT-5.4 xhigh by default) with no prior context to produce per-claim verdicts and action items.

Quick Start

Run the eeg-audit skill against your project directory after the analysis stages finish so it can generate an audit verdict and actionable fixes in audit-stage/.