Skip to content

Changelog

v0.2.0 (2026-03-17)

EEG consciousness research module — the first tool for measuring True Coherence in human brain activity.

EEG Module (pip install aime-loc[eeg])

  • Multi-Format Loading — EEGLAB (.set), EDF (.edf), BrainVision (.vhdr), BDF (.bdf), EGI (.mff), CSV, NumPy arrays via MNE backend
  • Consumer Device Presets — Built-in configurations for Muse 2/S, OpenBCI Cyton, Emotiv EPOC X, Neurosity Crown, g.tec Unicorn
  • Standard Preprocessing — Bandpass filter, notch filter, re-reference, EEG channel selection with full audit trail
  • PSD Epoch Extraction — Welch method with configurable duration, window, overlap, and frequency range
  • Server-Side TC Scoring — PSD sent to API; proprietary cognitive scoring performed entirely server-side
  • EEG Cognitive Profile — EEGCognitiveProfile extends CognitiveProfile with epochs, sampling rate, channel count, subject/task metadata
  • Multi-Subject Sessions — EEGSession container for batch processing with score_session(), summary tables, and CSV export
  • MNE Escape Hatch — to_mne() / from_mne() for custom MNE preprocessing (ICA, source localization, etc.)

EEG Visualization (pip install aime-loc[eeg,viz])

  • PSD Plot — Mean PSD with standard deviation shading, log scale
  • Time Series Plot — Power evolution across epochs with rolling trend and TC annotation
  • Cognitive Radar — 13-axis radar chart with journal presets (Nature, IEEE), multi-profile overlay
  • Scalp Topomap — Standard band power topographic maps via MNE (falls back to bar chart if no montage)

EEG API Endpoints

  • POST /v1/eeg/score — Score EEG PSD epochs for True Coherence
  • POST /v1/eeg/batch — Batch score multiple EEG recordings

Cross-Substrate Comparison

  • Compare human EEG and AI model profiles on the same 13-axis radar
  • Same proprietary scoring applied to both substrates
  • Publication-ready overlay figures

Tests

  • 58 new EEG tests covering models, scoring, spectrum, montage, visualization, and E2E pipeline
  • Synthetic EEG fixtures (sine waves at known frequencies) for deterministic ground truth
  • IP audit: verified no proprietary scoring algorithms or thresholds in shipped SDK code

Documentation

  • 9 EEG guide pages: Quick Start, Loading Data, Preprocessing, Scoring, Multi-Subject, Consumer Devices, Cross-Substrate, Real-Time, Visualization
  • 4 EEG API reference pages with mkdocstrings auto-generated documentation
  • 3 EEG examples: Quick Analysis, Research Study, Cross-Substrate Paper
  • Updated homepage and Getting Started with unified AI + EEG content

v0.1.0 (2026-03-16)

Initial release of the AIME LOC Python SDK.

Features

  • LOC Client: Sync (LOC) and async (AsyncLOC) clients with automatic retry, polling, and error mapping
  • 13-Function Cognitive Profiling: Full LOC V7 True Coherence scoring
  • Data Models: CognitiveProfile, ModelComparison, TrainingAudit, BenchmarkResult, Leaderboard
  • Visualization: 13-function radar charts, bar charts, benchmark heatmaps, delta charts
  • Publication Export: Journal presets (Nature, IEEE), PNG/SVG/PDF export, LaTeX tables
  • Data Export: JSON, CSV, LaTeX, markdown
  • MCP Server: 4 tools (scan_model, compare_models, get_leaderboard, training_audit)
  • Server-Side Coherence Computation: All cognitive scoring performed via the API
  • Known-Answer Tests: Verified against 5-model benchmark ground truth

API Endpoints

  • POST /v1/scan — Run model cognitive scan
  • POST /v1/compare — Compare two models
  • POST /v1/audit — Training audit
  • POST /v1/benchmark — Batch benchmark
  • GET /v1/leaderboard — Public leaderboard
  • GET /v1/models — List available models
  • GET /v1/usage — Check API usage

Benchmark Models (Ground Truth)

Rank Model TC Score
1 Llama-3.3-70B 15.37%
2 Mistral-Small-24B 11.46%
3 Qwen3.5-35B-A3B 10.24%
4 Qwen3.5-Distilled 9.07%
5 Gemma-3-12B 7.44%