Every paper entry (109 opened, 98 closed, grouped into about 31 independent signals) was tagged with trend alignment, momentum, volume, Bollinger %B and higher-timeframe conflict.
- 71 of 98 closed trades moved less than the 0.6% round-trip cost. The median holding time was 3 bars, and 37 trades closed within one bar.
- One strategy (local_mix) flipped direction far too often: 52 closed trades, 0 wins, about 40% of all losses.
- Trend-aligned, counter-trend, momentum-confirmed or not: every bucket lost money. Direction filters can't fix a cost problem.
- Suggestions passed to the code reviewer: an expected-move gate (≥2× costs), a cool-down before flipping direction, and multi-bar volume averages. Any rule change must be back-tested on longer history and counted per independent signal.
- The sample is tiny and heavily duplicated, so this is description, not statistics.