E-21. The same 6 families with long holds 6h/12h/24h + funding (2026-07-15)
- Preregistered D-027. The E-20 signal code was not changed; only the horizon was extended as a new lineage.
- All 36 runs completed. preliminary/train pass 0; val/test not opened.
- Best: majority 2d/20d/EMA60 + RSI reclaim + 24h. 369 signals, 0.423/day, gross +0.19040%, funding +0.01957% paid, baseline net -0.02917%, net at 1.5× costs -0.12917%. Gross positive in 4 of 6 windows and at 3 horizons.
- Pattern: all 6 majority-macro families had 24h as their best horizon, with gross +0.124~+0.190%, but all were negative after market-order costs + funding. Slow macro was weaker.
- Verdict D-028: official promotion refused. RSI 24h is not deleted; it is kept as a reference group for F1/F2 execution research, but it is not called alpha without real maker fills, non-fills and markouts.
- The cost counterfactual on the same maker-entry/taker-exit path is +0.0508%, but at 1.5x costs it is -0.0092%. This gap means strategy performance is very sensitive to real fill selectivity.
Source and migration evidence
Source: backtest/docs/experiment-log.md#E-21
Original SHA-256: 069c4aaaf53a20930ca8d94ba6365b4ce2b6bc560171d662a4c179ea55a6857c
Raw data and account information are not part of the public package. The original results do not replace statistical or scientific revalidation. The English text is a translation; numbers, tables and charts are carried over unchanged.