Evolutionary Trading Algo

Engine fix

Engine fix: Sharpe calculation FP-noise + deterministic-R guards

Three identical -1pct returns produced sd=1.3e-17 (FP rounding, not a real signal), giving Sharpe = -1.2e+16. The walk-forward aggregator averaged that into 9 digital-asset market OOS folds and reported a fleet-wide -1.35e+15. Root-caused, fixed, regression test added. Same iteration caught a second variant: 4 wins all hitting identical target-result wins produce sd/mean ratio ~1.15e-5 - also meaningless for Sharpe. Guard tightened to 1e-3 relative dispersion.

Public summary

Three identical -1pct returns produced sd=1.3e-17 (FP rounding, not a real signal), giving Sharpe = -1.2e+16. The walk-forward aggregator averaged that into 9 digital-asset market OOS folds and reported a fleet-wide -1.35e+15. Root-caused, fixed, regression test added. Same iteration caught a second variant: 4 wins all hitting identical target-result wins produce sd/mean ratio ~1.15e-5 - also meaningless for Sharpe. Guard tightened to 1e-3 relative dispersion.

Public record

  • Record type, publication date, and public-safe summary.
  • Aggregate evidence posture and operating boundary.
  • Whether the record belongs to research, runtime, safety, evidence, governance, or architecture.

Protected review

  • System names, market-specific routing, and implementation notes.
  • Protected dashboards, sensitive records, and control actions.
  • Detailed operator decision notes that could expose live decision mechanics.