Research
Research
TRAVIDENCE publishes its research in the open: the technical paper documenting the validation framework — the statistical methods we use to detect the signals of overfitting — and its executive brief. The English version is registered as a preprint on SSRN with a permanent DOI; both documents are free to download, in English and Spanish.
Technical paper — preprint
Audit-Grade Pre-Validation of Trading Strategies: Five Contributions to the López de Prado Stack
Ignacio Arias · Travidence, LLC · v1.0 · 2026
Pre-validation ahead of the prop firm challenge is a structural blind spot of contemporary retail trading. FPFX Tech data on 300,000+ accounts document that 86% of challengers fail to clear the evaluation and only 7% ever receive a payout — systematic evidence of overfitting that existing tools do not resolve at audit grade. Implementation libraries implementing the López de Prado stack (mlfinlab, RiskLabAI) operate as technical infrastructure without issuing certification; the aggregation-only retail consumer offering delivers an aggregate score without gates or procedural signing. Prop trading firms, simultaneously, sit on a structural conflict of interest that prevents them from offering honest pre-validation. This institutional vacuum motivates an independent third-party methodology with audit-grade discipline.
This paper presents five novel contributions to the López de Prado stack: (i) dual n_trials reporting in the Deflated Sharpe Ratio to taxonomize selection bias; (ii) Capital-normalized Calmar that strips margin-leverage inflation; (iii) sub-period schema with monotonic decay detection; (iv) component isolation test for multi-component strategies; (v) vol-on / vol-off MUST-PASS regime taxonomy. The five integrate into the Robustness Score — a hybrid continuous-gate score on [0, 100], inspired by the procedural discipline of the professional auditing industry.
The methodology is applied to three textbook strategies on public data: Golden Cross on CL crude oil futures (2000–2025), RSI-14 on BTC perpetual (2020–2025), and Asian Range Breakout on EUR/USD (2015–2025). None reaches a Robust band — consistent with the documented 86% failure rate. Under the production audit engine (trade-level returns, self-derived p50 regime cut, admissibility floors), CL and BTC close with a Disclaimer of Opinion for insufficient sample (18 and 31 trades), while EUR/USD emits the framework's public end-to-end score: 56 / 100, Limited Robustness, with no gate binding. The gate-vs-continuous behavior against pure aggregation that defines the defensible methodological differentiator of audit-grade pre-validation is exhibited by the repository's NQ Opening Range Breakout case — admitted with 2,424 trades and capped by the regime and subperiod gates; across the five repository case studies the portfolio covers the three public outcomes of the framework: score emitted, gated, and insufficient sample.
Preprint published on SSRN: https://doi.org/10.2139/ssrn.7308022
Download PDF (English) · Download PDF (Spanish)
How to cite
Arias, I. (2026). Audit-Grade Pre-Validation of Trading Strategies: Five Contributions to the López de Prado Stack. SSRN. https://doi.org/10.2139/ssrn.7308022
The Spanish PDF is the author's version of the same work; all citations point to the DOI.
Executive brief
Audit-Grade Pre-Validation of Trading Strategies — Executive Brief
The executive version of the same framework: the full argument, without the math.
Download PDF (English) · Download PDF (Spanish)
Apply the research
The two free tools on this hub apply, in your browser, two of the ideas the paper develops: Overfitting calculator → · Monte Carlo simulator →