Every strategy you'll see here passes the same six-step audit. This is the framework most retail "course gurus" have never heard of, and the reason their backtests don't survive contact with reality.
Each is taught in detail in the curriculum. Each must pass before a strategy goes live.
The headline Sharpe is a lie when you backtested 10,000 strategies. We deflate every reported Sharpe by the number of trials, autocorrelation, and skew/kurtosis (Bailey & López de Prado, 2014).
Your students stop being fooled by 3.0+ Sharpe screenshots and learn to ask: "how many trials produced this?"
k-fold cross-validation leaks information through label overlap. CPCV purges training folds of any sample whose label horizon overlaps the test set, with an embargo period to neutralize serial correlation.
Your students will stop overfitting and ship strategies that survive contact with reality.
Single backtests are overfit by construction. We use anchored walk-forward: re-estimate every N bars, validate on out-of-sample window, never look forward, never re-tune on the test set.
A simple rule your students can apply tomorrow: never report a metric that touched the test set.
Probability that the true Sharpe exceeds a benchmark, given sample length and return moments. Far more honest than a point estimate — especially below 36 months of live data.
Your students learn to say "73% confidence Sharpe > 1" instead of "my Sharpe is 1.4".
Linear slippage + impact-square-root model + bid/ask. Strategy survives a 2× cost-multiplier stress test or it does not deploy. Most retail "edges" die at this step.
Your students kill bad strategies in 5 lines of code, before risking a dollar.
Fractional Kelly with a posterior over the edge, not a point estimate. Shrinkage toward zero proportional to estimation uncertainty. The result: smaller bets when you know less.
Your students replace gut sizing with a rule that adapts to how confident the data actually allows them to be.
All performance figures derive from a read-only broker API and an independent fund administrator. Audit reports are available under NDA.
Every reported metric is deflated by the number of trials and adjusted for non-normal returns.
All datasets are point-in-time. Earnings, fundamentals, and corporate actions are timestamped at first public availability — never restated values.
Universe definitions include delisted instruments. Backtests run on the same survivorship-free universe as live.
Every strategy ships with a Docker image, frozen requirements, raw data hashes, and the seed used. Anyone with the package can rerun the experiment bit-for-bit.
We publish capacity estimates. When AUM approaches soft caps, the strategy closes. Educational content is not affected by this.
Our framework draws directly from these works. If you teach trading without having read them, you are guessing.
If half of these terms are new — read the free 12-page Primer first, then decide.