A retail quant trading daily bars with public data isn't going to out-compete Citadel. That's fine, because that's not how you lose. You lose by fooling yourself — by torturing data until it confesses, then trading the confession. So I built the lab around the failure mode instead of the upside.
The structure borrows from pod shops: five pods — trend, cross-sectional momentum, volatility premium, FX carry, and an incubator — each with a mandate, a risk budget, and kill criteria written down before testing begins. Under them, 64 hypotheses drawn from the academic literature: post-earnings drift, closed-end-fund discounts, the low-volatility family, VIX term structure, factor portfolios back to 1963.
The pipeline is the product
Every backtest ever run — including throwaways — gets a row in an append-only registry. 599 so far. That number isn't bookkeeping; it's the denominator. Deflated Sharpe ratios are computed against all 599 trials, because a “2.0 Sharpe” found on the 400th attempt is mostly luck. Rules freeze in writing before any formal test. Strategies get at most four free parameters. Each hypothesis may touch its holdout window exactly once, ever. And the standing prior: published Sharpe ÷ 2 — if you're beating the literature, find the bug.
The graveyard
The strategies that looked best in-sample died hardest. An open-gap reversal posted a 5.85 Sharpe in training — frozen validation returned −0.31, and with realistic costs, −2.57. Its sibling went 3.21 → −2.20. A crypto funding-rate carry showed a 7.28 Sharpe and was demoted anyway, because it rested on 129 observations. Of 58 graded hypothesis buckets, 31 failed despitea positive selected Sharpe — they just couldn't clear the multiple-testing bar.
One idea survived the full train → validation → holdout gauntlet: developed-markets profitability (RMW), holdout Sharpe 0.50. It was promoted cautiously — as an independent alpha source for further research, not a strategy. That ratio, 1 in 64, is the honest base rate of edge-hunting with public data. Most quant content sells you the 5.85. The lab exists to shoot it.