Gate Design and Stage-Dependent Incentives in Retail Proprietary-Trading Evaluations: Why passing is not standalone evidence of skill, and why the product fails to pay under measured trading constraints

Battle Count: 50

Directly relevant to quantitative trading in multiple dimensions: (1) Provides rigorous framework for evaluating trading strategies under realistic constraint geometry (trailing drawdown, consistency rules, minimum days); (2) Demonstrates that pass rates are manufacturable by position sizing alone, undermining naive skill inference from evaluation performance; (3) Quantifies the cost-cadence pincer that constrains viable strategy design; (4) Shows that contract geometry dominates strategy selection in evaluation products; (5) Provides first-passage tools applicable to any drawdown-constrained trading problem; (6) The zero-edge control methodology is directly transferable to any strategy evaluation pipeline; (7) The frequency asymmetry result (9x joint gate difference) has implications for any multi-stage trading evaluation. The paper is primarily about product design and contract economics rather than generating trading signals, but its findings are essential context for anyone evaluating or designing trading evaluation systems.

The core simulation (Monte Carlo over (μ,σ,f) surface, two-stage evaluation/funded modeling, first-passage calculations) is moderately complex but well-specified with fixed parameters. The research protocol with five sequential gates, pre-specified kill criteria, and zero-edge control adds methodological complexity. The adversarial test, hedged-pair simulation, and seller-side model each add distinct implementation challenges. The multi-agent orchestration layer (8 operational agents) described in Section 2 is the most complex component but is noted as non-operational at time of writing. Reproduction requires numpy with specific versions for exact seed matching. The paper provides a GitHub repository with all code, reducing practical implementation burden significantly.

All simulation code, figure-generation scripts, and build script are public at https://github.com/nicholasbhall/gate-design-prop-evals. Fixed seeds used throughout. Detailed verification protocol in Appendix A specifies exact reproduction checks including analytic anchors, cross-validation against independent backtest engine, and adversarial test reproduction. 349 full agent run transcripts retained. Every reported figure traces to a retained transcript. Pre-specified kill criteria, fixed split date (7 September 2021), and documented protocol corrections all recorded. Citation discipline explicitly stated with every reference verified individually.