Skip to content

The Learn tab

The Learn tab is a structured, hands-on course on the methodology Holdout Lab exists to teach: how to tell a real edge from an overfit one. It’s the front door for newcomers — you can take it without writing any code, and the exercises run real backtests and the real gauntlet on the app’s own engine, not canned demos. Progress is tracked locally as you complete lessons.

Four modules, worked in order:

  1. Foundations — What a backtest is · Reading the results · Signal vs noise. What the engine actually does and how to read a result without fooling yourself.
  2. The overfitting trap — What overfitting is · In-sample vs out-of-sample · Why a beautiful backtest can lie. The core failure mode the whole product is built to catch.
  3. The gauntlet — Does it beat random? (the rule test) · The held-out test · Walk-forward optimization · Deflated Sharpe · Monte Carlo robustness. Each stage, and what it protects against — the hands-on companion to the gauntlet.
  4. Forward-test & graduate — Why paper trade · From a template to your own idea · You’ve graduated. Turning a survivor into a forward test and then your own strategy.

The Learn tab includes a small library of curated, well-understood strategies — ibs_reversion, turn_of_month, sma_cross, rsi_pullback, donchian_breakout, tsmom — that you can run straight from the lesson with two buttons:

  • Quick check → POST /api/learn/quick-check runs just the rule test (the fast “does the entry beat random timing?” reveal).
  • Run gauntlet → POST /api/learn/run-gauntlet runs the full ladder (optionally reusing the quick-check’s rule test so it isn’t redone).

Both enqueue the same run_ladder_job the rest of the app uses, so results land as ordinary BacktestRun rows and roll up into the Research ladder’s readiness score — the Learn tab and the Research tab are looking at the same honest scoreboard.

The graduate step points you at the Studio (describe your own idea in plain English) and the strategy contract (write one as code), then into the full research workflow. What survives can be promoted to paper trading.