This study audits whether candle-based machine learning models can translate predictions of cryptocurrency extrema or short-horizon outcomes into positive Binance Spot trading strategies. Numerical results stem from scripted fixed-seed model runs and deterministic simulators; human-supervised AI agents supported the July 20 evidence integrity revision through literature retrieval, separate critique tasks, artifact reconciliation, documentation, and source packaging, rather than trading decisions.
The strongest later-period evidence indicates, after extensive predecessor search, that an unchanged ten-pair mandatory daily selector lost 6.72% over 19 July cycles at an assumed 31-bps completed-cycle cost, achieving only 3 wins and 16 losses.
In short model-specific July evaluations, the validation-selected local-minimum policy returned -1.79%, while the local-maximum sell-to-cash/re-entry policy underperformed continuous holding by 2.80%; their gross mean advantages of 11.11 and 12.21 bps fell below even the 21-bps stress test.
A Gurgul-inspired, OHLCV-only daily adaptation achieved minimum/maximum ROC AUC of 0.874/0.896 but average precision of only 0.134/0.116, losing 44.30% over seven cycles compared to -41.20% for buy-and-hold.
A forensic audit also downgraded an earlier One4All '30-day holdout': its dates influenced prior architecture work, its four-hour outcome horizon was not purged at split boundaries, it used same-close entry, and its raw result directories were absent.
Across the tested, mostly exploratory protocols, event-ranking performance did not establish positive executable strategy value. Every operational decision remains NO_TRADE.
Blogger's Review: This article reveals the limitations of candle-based machine learning models in real trading through rigorous auditing, emphasizing the significance of AI in decision support while warning investors against over-reliance on model predictions.