https://backtestdiscipline.com/Published trading strategies, independently tested. Evidence-backed evaluations and practical research lessons by Mikhail Klemin.en-USThu, 17 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/research/research-007/Thu, 17 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/research/research-007/We converted Data Trader's 5-minute Bitcoin range-fade rules into a bounded deterministic specification before evaluation. On one year of complete post-publication BTCUSDT data, the implementation produced 739 trades and failed its predeclared survival rule.https://backtestdiscipline.com/notes/why-positive-annualized-return-still-failed/Mon, 14 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/notes/why-positive-annualized-return-still-failed/A positive return describes one dimension of a backtest. It does not override a multi-criterion acceptance rule that was fixed before the result was known.https://backtestdiscipline.com/research/published-strategy-replication-001/Mon, 14 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/research/published-strategy-replication-001/We tested all four published BTC trend-following horizons on a subsequent out-of-sample period. Every variant reduced maximum drawdown, but every mean excess-return estimate was negative. Under the frozen family rule, 0 of 4 variants survived.https://backtestdiscipline.com/notes/when-prose-and-code-disagree/Sat, 05 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/notes/when-prose-and-code-disagree/Benchmark 004 reproduced an external channel-breakout notebook exactly enough to expose a harder question: when the written explanation and the executed program differ, which object are you actually testing?https://backtestdiscipline.com/research/research-005/Sat, 05 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/research/research-005/The final experiment returned +5.26% net over the frozen holdout and a +20.50% annualized arithmetic mean return. It was still classified FALSIFIED because the Sharpe and dependence-aware confidence criteria did not pass.https://backtestdiscipline.com/research/benchmark-004/Sat, 05 Sep 2026 00:00:00 +0000https://backtestdiscipline.com/research/benchmark-004/Benchmark 004 asked a deliberately narrow question before any prospective extension: could independent machinery recover the same executable strategy behavior and the same underlying market-data panel? The terminal answer was FULL_REPLICATION_MATCH.