Interactive strategy research lab for retail investors to understand risk, overfitting, and regime dependence.
Quant Lab is not a stock predictor or trading bot. It is a local-first research cockpit for turning a strategy hypothesis into a backtest, stress test, robustness review, and research memo.
Most retail backtest tools make the headline return easy and the fragility hard to see. Quant Lab does the opposite: every result keeps assumptions, data provenance, drawdowns, costs, parameter sensitivity, start-date dependence, and failure modes close to the chart.
- Create or open an experiment.
- Run a backtest.
- Open
Resultsto inspect metrics, drawdowns, fills, warnings, and deterministic review. - Open
Robustnessto run cost, start-date, and parameter sensitivity. - Export the Markdown tear sheet or write a wiki summary.
| Area | Status | Notes |
|---|---|---|
| Experiment cockpit | Working | Create/import/export experiments with hypothesis, universe, dates, costs, cash policy, benchmark, and execution timing. |
| Strategy templates | Working | Buy and Hold, Moving Average Filter, Momentum Rotation. |
| Rule workbench | Working | Edit executable strategy blocks and inspect compiled JSON. |
| Backtest engine | Working | Deterministic target-weight interpreter, rebalance simulation, costs, slippage, fills, equity, drawdown, benchmark curve. |
| Diagnostics | Working | OOS analysis, rolling metrics, named regimes, portfolio risk, data reliability, provenance, deterministic quant review, bootstrap stress. |
| Robustness lab | Working | Cost sensitivity, start-date sensitivity, parameter sensitivity, and fragility verdict. |
| Research exports | Working | JSON export, Markdown tear sheet, wiki experiment summary, open question capture. |
| Portfolio polish | Working | README screenshot, feature matrix, known limits, agent docs, wiki updates. |
| Tests | Working | Python backend tests, TypeScript/Vite production build, frontend smoke script. |
React/Vite UI
-> src/api/experiments.ts
-> FastAPI backend
-> local JSON experiment store
-> yfinance market data cache
-> deterministic strategy interpreter/backtest engine
-> diagnostics, robustness, exports, wiki outputs
Core backend modules:
backend/quant_lab/api.py- FastAPI routes and dependency injection.backend/quant_lab/domain.py- dataclasses, enums, invariants, serialization.backend/quant_lab/programs.py- executable strategy blocks and interpreter.backend/quant_lab/engine.py- backtest orchestration.backend/quant_lab/robustness.py- cost/start-date/parameter sensitivity.backend/quant_lab/wiki_exports.py- tear sheets, wiki summaries, open questions.
- Local-first app: no auth, multi-user storage, or hosted deployment model.
- Market data uses yfinance cache; data quality and survivorship limits remain visible assumptions.
- Execution model is simplified: no taxes, borrow fees, liquidity constraints, partial fills, or broker routing.
- Custom rules are editable as executable blocks, but not yet a full natural-language strategy builder.
- Research memo export is Markdown, not polished PDF/share page.
This repo includes an LLM-maintained wiki pattern:
raw/stores immutable source files.wiki/stores generated markdown pages.wiki/index.mdis the content catalog.wiki/log.mdis the append-only activity timeline.AGENTS.mddefines ingest, query, and lint workflows for future agent sessions.
Agent-readable technical documentation lives in docs/.
docs/index.mdis the docs catalog.docs/llms.txtis the compact agent manifest.
Edit .env if you want custom ports.
Backend:
python -m uvicorn backend.main:app --host 127.0.0.1 --port 8011Frontend:
pnpm dev -- --host 127.0.0.1 --port 5173Verification:
python -m pytest
pnpm.cmd build
pnpm.cmd smoke:frontendsmoke:frontend expects backend and frontend already running at default local URLs.