Outcome
Publisher stats match; catalog admission does not.
The frozen public-data replica reaches Sharpe 1.445, excess return 14.90%, volatility 10.31%, drawdown −19.68%, and correlation-to-beta 0.464 on the approximate 2001–2026 gross window versus publisher targets 1.44 / 15.8% / 11% / −18% / 0.49. Every fidelity check passes.
The untouched 2023-01-03..2025-12-31 base-cost holdout does not: Sharpe improvement versus monthly static beta is −0.64 against a +0.20 floor, and drawdown improvement is only +3.4pp against an +8pp floor. Adjacent parameters pass 0/5; extra lag and 2x costs stay negative. G13 decision is fidelity_only.
Why no Other / Strategies row
research.strategy_evidence.admission admits only promote. This run is fidelity-only research evidence. Issue acceptance asked for an Other / Strategies page, but publishing a catalog strategy after failed economic floors would be misleading. Follow-up #1321 tracks a later revisit if floors clear without retuning.
Artifacts
- Spec / protocol / evaluation / evidence:
research/findings/specs/prometheus-basic-trend.json,research/findings/prometheus-basic-trend-protocol.json,research/findings/prometheus-basic-trend-evaluation.json,research/strategy_evidence/prometheus-basic-trend.json - Implementation:
backtest/prometheus_basic_trend.py,research/prometheus_basic_trend_runner.py - Experiment:
research/experiments/1319.json,research/experiments/1319.ipynb,docs/experiments/1319-analysis.html
Limitations
Inferred estimators, proxies, costs, lag, and the 3x cap are explicit QuantModel choices frozen before returns. BTC PriceUSD is non-executable. High turnover makes net results cost-sensitive. No post-result parameter repair was applied.