Real-time collaboration health monitoring engine for software engineering teams.
DEFERRED UNTIL POLYEDGE FINISHES SHIPPING
Ingests project management board activity, git history, and status reports. Computes collaboration metrics (Gini, entropy, bus factor, churn decomposition) segmented by work type via zero-shot NLI classification. Surfaces findings through a live web dashboard with per-sprint trend analysis.
v2 adds: NLI-enhanced commit/card classification (DeBERTa-v3-base), hot/cold path architecture, multi-signal fusion, circuit breaker resilience, and academically structured hypothesis testing. See docs/ for methodology.
v1 forensics engine archived in v1/.
pip install -r requirements.txt
# Optional: download NLI model (~440MB, falls back to keyword+diff if absent)
python scripts/download_models.py --verify
# Configure team
cp reconcile/config_template.py reconcile/config_local.py
# Edit config_local.py
# Start server
python -m reconcile.main --serve
# Dashboard at http://localhost:8080See reconcile/README.md for full documentation.
python -m pytest reconcile/tests/ -v # 286 tests