Autonomous proof engine generating seven layers of verified training data.
npm install
cd sidecar && pip install -e ".[dev]" && cd ..
npm run sidecar:dev # Terminal 1
npm run tauri:dev # Terminal 2This repository follows Gitflow plus mandatory TDD phase gates for every code change.
- Long-lived branches
main: production-ready, release-tagged historydevelop: integration branch for completed feature work
- Working branches
feature/*: branch fromdevelop, merge back todevelopvia PRrelease/*: branch fromdevelop, merge tomainand back-merge todevelophotfix/*: branch frommain, merge tomainanddevelop
- Required lifecycle for all code changes
- Red → Green → Blue is mandatory
- No direct implementation bypassing Red evidence
git checkout develop
git pull origin develop
git checkout -b feature/<task-id>-<short-kebab-description>
git push -u origin feature/<task-id>-<short-kebab-description>
gh pr create --base develop --head feature/<task-id>-<short-kebab-description> --fillNo implementation before failing tests (Red)
Minimal implementation to pass tests (Green)
Refactor only after green tests (Blue)
PRs without Red evidence are non-compliant
- Rust Backend (Tauri 2): Loop engine, orchestrator, critic, worker pool, SQLite
- Python Sidecar (FastAPI): SymPy, Pint, Lean validators + Council, Scout, Librarian agents
- React Frontend: Zustand stores, proof tree, agent dashboard, training data stats
| # | Source | Signal |
|---|---|---|
| 1 | Solvers | Step-level PRM |
| 2 | Solvers | Full trajectories |
| 3 | Orchestrator | Routing decisions |
| 4 | Council | Metacognitive deliberation |
| 5 | Scout | Research query-result-outcome |
| 6 | Critic | Prediction vs truth |
| 7 | Librarian | Curation traces |
License: Proprietary - Vario Automation