cyb50-quant/research/dragon/v2/dragon_deep_oversold_selective_veto_review.md

Dragon Deep Oversold Selective Veto Review

Scope

• Alpha-first research branch only.
• Formal workbook-preserving baseline remains unchanged.
• Tested whether narrow local veto rules can outperform broad weak-subtype blocking.

Main Result

• Baseline: 107 trades, avg_return 2.52%, profit_factor 3.31, real BUY / SELL overlap 106 / 105
• Best narrow branch: selective_veto_plus_mixed_c1_lt_10_2_no_ql
• Result: 103 trades, avg_return 2.81%, profit_factor 3.96, real BUY / SELL overlap 102 / 101
• Broad comparison branch: block_all_remaining_weak_subtypes
• Result: 102 trades, avg_return 2.78%, profit_factor 3.76, real BUY / SELL overlap 101 / 100

Why The Best Narrow Branch Wins

• It removes four clearly weak deep-oversold trades:
• 2022-02-09 -> 2022-02-11 mixed_oversold
• 2022-03-11 -> 2022-03-15 positive_b1_rebound
• 2024-01-18 -> 2024-01-22 shallow_false_start

• 2024-01-23 -> 2024-01-29 shallow_false_start

• It keeps these borderline or partially useful trades that broad blocking would also kill:

• 2022-03-16 -> 2022-03-25 positive_b1_rebound

• 2024-08-29 -> 2024-09-09 shallow_false_start

• Quant implication:

• narrow veto preserves more surviving edge than broad subtype deletion

• the weak-family problem is not “all positive-B1” or “all shallow false starts”

• the weak-family problem is a smaller subset of locally fragile oversold rebounds

Candidate Alpha-First Veto Rules

• Positive-B1 veto:

• block positive_b1_rebound when c1 < 15.3

• Shallow false-start veto:

• block shallow_false_start when c1 >= 12 and b1 > -0.025

• Mixed oversold veto:

• block mixed_oversold when c1 < 10.2 and there is no same-day QL

Quant Judgment

• This is the first deep-oversold alpha-first branch that is clearly better than a broad weak-subtype block.
• It improves avg_return and profit_factor more than block_all_remaining_weak_subtypes, while losing fewer aligned dates.
• Therefore the preferred Track-A direction is now:
• do not start from broad subtype deletion
• start from narrow pathological-pattern veto

Decision

• Workbook-preserving baseline: no change.
• Alpha-first branch: the best next implementation candidate is selective_veto_plus_mixed_c1_lt_10_2_no_ql.