Abstract
Praxis is a library of audited systematic trading strategies. Published claims overstate themselves, routinely and measurably [2, 3]. We read ≈480 new papers a month, 42,984 strategy papers mapped to date, and rebuild the most consequential to the paper’s own spec. Each is re-run on bias-free data, out of sample, net of all real costs, across crisis regimes, accounting for overfitting [4] and luck [1]. Every claim leaves validated, overstated, or fails, and its code integrates into IBKR, Bloomberg or LEAN.
1Findings
| # | Paper | Asset | Claimed | Measured | Claimed → Measured | Verdict |
|---|---|---|---|---|---|---|
| 1 | The Volatility Edge — dual VIX-ETN signal | Vol | Sharpe 1.00 | 1.03 | validated | |
| 2 | Network momentum across asset classes | X-asset | Sharpe 1.51 | 0.62 | overstated | |
| 3 | Lazy factor + probability weighting | Gold·BTC | $136,216 | $4,374 | fails | |
| ⋮ | More papers can be viewed in Your Library → | |||||
2Methodology
Definition 1 (Measured edge). The measured edge is the paper’s own headline statistic, recomputed from a full re-implementation under three conditions: (i) out of sample, (ii) net of real costs, (iii) on survivorship-clean, point-in-time data. Table 1 reports it beside the claim.
Theorem 1 (Selection). Among N backtests of noise, the expected best Sharpe ratio grows as √(2 ln N) [1]. Publication reports the maximum, so a claimed edge exceeds the true one in expectation. The excess is identified only by re-measurement under Definition 1, and Eq. (1) turns the gap into the verdict.
Remark. The verdict is computed, not argued.
3Access
| Reader — free | Praxis Pro — £29/mo | |
|---|---|---|
| A library of clustered systematic trading strategies | 5 papers | 42,984 |
| Strategy validation reports under live trading conditions | 5 | all |
| Reproduction code, integrated with Bloomberg, IBKR, etc. | 5 | all |
| Integration with your AI agent — Claude, ChatGPT, Cursor | 5 | full corpus |