A walk-forward, anti-mirage study of systematic crypto strategies — in which our own champion passes every readiness check, and our own gate proves it was luck. Five years of real data. 53 unseen folds. No excuses.
We tested a family of systematic strategies — momentum, mean-reversion, breakout, factor-neutral, and delta-neutral funding carry — on BTC, ETH and SOL: ~5 years of daily Coinbase data, strict walk-forward validation (train on 180 bars, embargo 3, grade only on the next unseen 30), 53 independent folds, and realistic costs of 12 bps per position flip. Passing our 7-check readiness gate requires: average out-of-sample Sharpe ≥ 0.35, ≥55% of folds profitable, beating buy-and-hold, Monte-Carlo blow-up risk ≤ 10%, survivable drawdowns, low turnover, and ≥ 20 folds of evidence.
Then we go further — because a 7/7 can still be a lie. The anti-mirage championship gate adds the Deflated Sharpe Ratio (Bailey & López de Prado 2014), the Probability of Backtest Overfitting via CPCV (Bailey et al. 2017), and cross-asset generalization. Ten checks. The last three exist to catch us.
SOL's directional champion, a regime-gated trend strategy, did everything right:
Under most published methodology, this gets funded. We sent it to the judges instead.
Accounting for the ~40 configurations tried during selection, the apparently excellent Sharpe is indistinguishable from noise, and cross-validated combinatorics say there is a 79.3% probability the backtest is overfit. The 7/7 was forty trials' worth of luck wearing a championship belt. A base readiness gate is a necessary filter, not a sufficient verdict — this single result is the strongest argument in this paper, and it is an argument against ourselves.
The durable edge is delta-neutral perpetual funding carry — long spot, short perp, collect the funding premium — and it is asset-specific. ETH requires its native staking yield (long-stETH "LST-basis carry") to clear the gates.
| ASSET | DIRECTIONAL (BEST) | CHAMPION CONFIG | /10 GATE | LIVE CHECK |
|---|---|---|---|---|
| SOL | 7/7 · Sharpe 0.91 | Gated Trend — mirage | 8/10 · DSR 0.029 · PBO 79.3% | … |
| BTC | ≤ 3/7 | META-ALLOC carry+trend | 10/10 · DSR 0.996 · PBO 1.5% | … |
| ETH | ≤ 4/7 | SAM+META LST-basis · Sharpe 4.61 | 10/10 · DSR 1.00 · PBO 2.4% | … |
We instantiated two further agents with opposite personalities — Sam (aggressive momentum) and Lianna (patient mean-reversion) — and ran the identical gauntlet. The mirage pattern replicated: Sam's 7/7 also failed anti-mirage (DSR 0.354, PBO 39.2%). Then we gave every agent access to a shared house library of gate-cleared strategies. In a single run:
The weakest agent was lifted by knowledge transfer, not retraining — she kept her own style and inherited the house's proven carry engine. We propose this as a reusable framework: competition on identical unseen data exposes mirages; gate-cleared strategies propagating as shared infrastructure compound honest edge.
BTC's champion survives doubled costs (24 bps, still 6/7), 30%-thinner funding, all 9/9 parameter-sweep combinations (a plateau, not a knife-edge), and an inverted-funding regime (the sleeve flips to short-carry and still earns). Cost cliff ≈ 30 bps/flip — 2.5× margin over modeled costs. ETH holds 6/9 sweeps with a ≈ 24 bps cliff. The edge is regime-dependent: one of three sub-periods (compressed funding) turns negative for BTC — we say so, because it's true.
The backtest models funding income, flips and a 90% capture haircut — but not per-venue fees, slippage, entry/exit basis, LST depeg, or smart-contract risk. Avg-fold Sharpe is inflated by short-window annualization; trust the overall OOS figure. The /10 gate is the strongest in-sample proxy, but the definitive test is the live forward track, which runs daily and publicly. Nothing here is funded at scale. Nothing here is financial advice. If our numbers move, this page will tell you before we do.
node training-camp/funding-fetch.mjs ETH|BTC|SOL node training-camp/staking-fetch.mjs node training-camp/gauntlet.mjs daily BTC --stress # → docs/CHAMPION_BTC_DAILY.json npm run stable # → the whole roster, one command
Live dashboards: research · the gym · the league. Raw records ship with this site — open them.
Bailey & López de Prado (2014), The Deflated Sharpe Ratio · Bailey, Borwein, López de Prado & Zhu (2017), The Probability of Backtest Overfitting · Blitz, Huij & Martens (2011), Residual Momentum · Schmeling, Schrimpf & Todorov (2024), Crypto Carry, BIS WP 1087 · He & Manela (2022), Fundamentals of Perpetual Futures.