Evidence Sync

Personal evidence engine — autonomous pipeline that monitors PubMed for new research, extracts study data via Claude, and maintains living meta-analyses with drift detection.

Built for tracking scientific evidence on health topics that matter to you: supplements, nutrition, metabolic health, sleep. Turns static medical reviews into living, versioned datasets.

How It Works

PubMed API → Study Finder → Claude Extractor → Statistical Engine → Alerts
                                    ↓
                              Git Dataset (YAML) → Drift Detector

1. Monitor — Scheduled PubMed search for new RCTs matching your review topics
2. Extract — Claude reads abstracts and extracts structured data (effect sizes, CIs, bias)
3. Analyze — Random-effects meta-analysis (DerSimonian-Laird) with heterogeneity metrics
4. Detect — Compare against previous analysis, alert on significant evidence drift
5. Report — Forest plots, funnel plots, PRISMA 2020 compliance

Quick Start

pip install -e .

# Initialize a review topic
evidence-sync init protein-muscle-preservation \
  --name "Protein intake and muscle preservation in aging males" \
  --query "protein intake muscle preservation aging male RCT" \
  --effect-measure standardized_mean_difference \
  --outcome "Lean mass retention"

# Run the full pipeline
evidence-sync run protein-muscle-preservation

# Or step by step
evidence-sync search protein-muscle-preservation
evidence-sync extract protein-muscle-preservation
evidence-sync analyze protein-muscle-preservation
evidence-sync report protein-muscle-preservation

Requirements

• Python 3.11+
• ANTHROPIC_API_KEY environment variable
• Internet access (PubMed API)

Validation

Validated against Cipriani et al. 2018 Cochrane review (SSRIs for depression) — pooled effect within expected range. 128 tests (119 unit + 9 integration).

License

MIT