Dragon Glued Refine Experiments

Baseline branch: alpha_first_selective_veto.
Goal: verify whether the hot glued veto can be narrowed after attribution without giving back too much trade quality.

Summary

baseline_alpha_first: trades 103, avg_return 2.81%, profit_factor 3.96, short_avg_return -1.15%, 00-05d -1.72%, 06-10d -0.69%, real BUY / SELL 102/101
glued_veto_low_weak_range: trades 99, avg_return 3.02%, profit_factor 4.38, short_avg_return -1.06%, 00-05d -1.61%, 06-10d -0.69%, real BUY / SELL 98/97
glued_veto_hot_and_low: trades 92, avg_return 3.35%, profit_factor 4.95, short_avg_return -0.99%, 00-05d -1.67%, 06-10d -0.58%, real BUY / SELL 90/89
glued_veto_hot_cap75_and_low: trades 92, avg_return 3.36%, profit_factor 4.98, short_avg_return -1.00%, 00-05d -1.67%, 06-10d -0.59%, real BUY / SELL 91/90

glued_veto_low_weak_range: delta_avg_return 0.21%, delta_profit_factor 0.42, delta_short_avg_return 0.08%, real BUY / SELL 98/97
glued_veto_hot_and_low: delta_avg_return 0.54%, delta_profit_factor 0.98, delta_short_avg_return 0.16%, real BUY / SELL 90/89
glued_veto_hot_cap75_and_low: delta_avg_return 0.55%, delta_profit_factor 1.02, delta_short_avg_return 0.15%, real BUY / SELL 91/90

glued_veto_low_weak_range is the clean conservative upgrade candidate if governance still prioritizes overlap preservation.
glued_veto_hot_and_low remains the strongest quality-improvement branch but may still be too aggressive on overlap.
glued_veto_hot_cap75_and_low specifically tests whether the only super-hot positive sample can be restored without giving back too much of the glued cleanup benefit.
Current result is stronger than expected: glued_veto_hot_cap75_and_low dominates the old full glued candidate on both quality and overlap.
Refined-vs-full trade diff is minimal and interpretable: it restores 2021-11-05 -> 2021-11-18 and removes the fallback reroute 2021-11-22 -> 2021-11-30.
Candidate snapshot file: dragon_glued_refined_candidate_config.json.