Finding (replay of 2026-09-06)
Since #88 the backtest gaps a fill above the row's max_buy (trigger + 5 %, further capped at MAX_BUY_RISK_MULT x planned risk) instead of entry + 5 %. On the year-long tuned replay (250 sessions to 2026-09-04, horizon 60) that turned 0 gaps into 9-11 and moved mean R from +0.34 to +0.27/+0.28, independent of the multiplier (1.5 / 2.0 / 3.0 / off all land within one gap of each other; runs 34023518929, 34025638457, 34025639449, 34025637244).
So about ten fills a year open between trigger + 5 % and entry + 5 %, on breakouts whose confirming close was already above the trigger, and those ten averaged roughly +1.25 R: the best group of trades in the replay. The report's "do not chase above Max buy" rule refuses exactly them.
Why nothing changed yet
Ten trades, one year, a momentum regime. The trigger-based Max buy is the documented rule and is consistent with the scanner's own runaway test on closes. Loosening it on this sample would be curve-fitting.
What would settle it
- Re-run the comparison when a second year of data exists, or on the 2024-25 window (
--period 3y, --days 500) to see whether the gap-up band stays positive outside a momentum year.
- If it does, either base Max buy on the entry (breakout close) rather than the trigger for confirmed rows, or report both levels and let the reader choose.
Related: #81 (stop basis), #82 (score calibration).
Finding (replay of 2026-09-06)
Since #88 the backtest gaps a fill above the row's
max_buy(trigger + 5 %, further capped atMAX_BUY_RISK_MULTx planned risk) instead of entry + 5 %. On the year-long tuned replay (250 sessions to 2026-09-04, horizon 60) that turned 0 gaps into 9-11 and moved mean R from +0.34 to +0.27/+0.28, independent of the multiplier (1.5 / 2.0 / 3.0 / off all land within one gap of each other; runs 34023518929, 34025638457, 34025639449, 34025637244).So about ten fills a year open between trigger + 5 % and entry + 5 %, on breakouts whose confirming close was already above the trigger, and those ten averaged roughly +1.25 R: the best group of trades in the replay. The report's "do not chase above Max buy" rule refuses exactly them.
Why nothing changed yet
Ten trades, one year, a momentum regime. The trigger-based Max buy is the documented rule and is consistent with the scanner's own runaway test on closes. Loosening it on this sample would be curve-fitting.
What would settle it
--period 3y,--days 500) to see whether the gap-up band stays positive outside a momentum year.Related: #81 (stop basis), #82 (score calibration).