The research layer of a live, real-money options trading system: it discovers, tests and retires equity and options signals with point-in-time discipline. This repository is a showcase: it documents the design, the gates and real outputs. The engine source is private.
Live with real capital. The trading system this layer feeds places real orders with my own money every market day. Research and execution are kept apart: this layer has no order routing of its own, and every published signal is also settled against real market quotes in a shadow ledger, so research is graded separately from execution.
| Layer | What it is |
|---|---|
| Factor workshop | Proposes factors, compiles them in a sandbox, evaluates them, and keeps a lineage tree of every attempt, including the dead ones and why they died |
| IC evaluator | Daily cross-sectional Spearman IC against h-day forward returns, with t-stat, IR, hit rate and 20-day rolling IC |
| Faultline scanner | Scores where an options-driven price dislocation is building (gamma flip, open-interest walls, gaps, IV term-structure inversion) |
| Shadow ledger | Settles every published card against real quotes at three horizons, so the system grades itself |
Stack: Docker Compose · FastAPI (API + worker) · React/Vite dashboard with three tabs: Ops, Alpha Workshop, Particle Field.
- Gates before compute. A factor with no written hypothesis is dead on arrival. Fewer than 60 observations means it is recorded only, never promoted. At most 12 live factors at any time.
- Point-in-time or nothing. Fundamentals become usable on the SEC
acceptedDate+ 1 day. Snapshots are append-only. Options open interest is treated as T+1. - Honest settlement. Close-to-close returns and option-mid settlement are kept as separate tracks so one cannot flatter the other.
- Self-grading. The shadow ledger caught the first card's suggested entry at $1.04, about 12% above the actual mid. That gap is now a tracked metric instead of a hidden cost.
- docs/ARCHITECTURE.md: services, data flow, sandbox
- docs/METHODS.md: factor gates, IC evaluation, test constitution, faultline heat
- docs/SAMPLE_OUTPUT.md: output shapes and the first shadow-ledger finding
Research layer of a live system; execution trades real money and stays private. Data: a paid fundamentals/price feed plus an options data feed (v3 API).
Results before theory. Time-to-demo is 90 seconds: if it runs, you can see it. Open an issue to ask for a live walkthrough.
The engine, the factor definitions, the live ledger and the execution system. Happy to go deeper on tuning and evaluation details. A more aggressive causal-patch approach exists and is not public yet.
See NOTICE. Documentation shared for review. The engine source is not included and not licensed.