Learn

The Atlas Method

The concepts behind proving whether a trading edge is real.

Many trading strategies that look good in a backtest aren't robust. They're accidents that survived because nobody checked hard enough.

The concepts below are the checks that separate the two. They are not obscure — they are standard statistical practice in every field that deals with noisy data — but they are routinely skipped in retail trading, because skipping them produces better-looking results.

Filter by

01

Is your strategy fitting noise?

  1. 01Atlas method

    Backtest overfitting

    The problem: when a strategy has learned the noise in your historical data rather than a real pattern.

    Atlas check: Walk-forward validation and the multiple-testing penalty expose results shaped around historical noise.

    Learn about backtest overfitting
  2. 02Harness 04

    Multiple-testing bias

    The problem: why the twentieth idea you test needs to clear a higher bar than the first.

    Atlas check: The statistical bar rises with the number of configurations you have actually searched.

    Learn about multiple-testing bias
02

Does it survive unseen data?

  1. 03Harness 01

    Out-of-sample testing

    The problem: checking a strategy against data it was not built on.

    Atlas check: A held-back part of history is never used to build or tune the strategy.

    Learn about out-of-sample testing
  2. 04Harness 02

    Walk-forward validation

    The problem: testing a strategy the way it would actually have been used: built on the past, run on what came next.

    Atlas check: The strategy is rebuilt and retested across successive windows in chronological order.

    Learn about walk-forward validation
03

Is the signal actually doing anything?

  1. 05Harness 03

    Placebo control

    The problem: checking whether your entry signal is doing anything a coin flip would not.

    Atlas check: Random entries with matching risk, costs and exits test whether the entry signal adds anything.

    Learn about placebo control
  2. 06Harness 03

    Monte Carlo simulation

    The problem: testing your strategy against thousands of random alternatives to see if the edge is real.

    Atlas check: Every rigor check compares the strategy against thousands of seeded random-entry trades with matched risk and costs.

    Learn about monte carlo simulation
  3. 07Atlas method

    Look-ahead bias

    The problem: when a backtest quietly uses information that would not have existed yet.

    Atlas check: Builder blocks restrict decisions to information that genuinely existed at that moment.

    Learn about look-ahead bias
These aren't optional checks

The Atlas rigor harness

Every one of these is applied automatically to every strategy built on Atlas Edge.

See how the Rigor Harness works