Sandboxes.
Two browser-side toys for showing the mechanics of an ML / backtest loop. Both generate their own synthetic data and run the entire pipeline client-side — no backend, no real prices. For a backtest that runs on the platform engine against real predictions, see /backtest.
microstructure · 1 instrument
Synthetic order book (1–5 levels), 7 microstructure features including book-pressure that's silent at L1, ridge regression refit per walk-forward fold, leakage toggle. Shows what a tick-level ML pipeline looks like at the seam.
portfolio · cross-sectional
Synthetic prices for 8–60 tickers, momentum score on a closed lookback window, long the top quantile and short the bottom, rebalance every N bars, eat turnover-weighted costs. The price generator plants a slow drift so momentum picks it up — mechanics-only.
why two pages?
They show different things. The microstructure page is about single-instrument feature engineering and walk-forward refit — where leakage lives, why book-pressure earns its keep over L1. The portfolio page is about ranking across a universe and holding for N bars — where turnover and quantile sizing dominate. Neither says anything about a strategy you could actually trade; both make the loop legible.