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

Dragon Alpha Branch Governance

Scope

• Current formal alpha branch: alpha_first_selective_veto.
• Leading candidate branch: alpha_first_glued_refined_hot_cap.
• Goal: decide whether the refined branch should replace the current formal alpha branch or remain a governed candidate.

Headline Metrics

• current alpha: trades 102, avg_return 2.86%, profit_factor 4.04, real BUY / SELL 101/101
• refined candidate: trades 91, avg_return 3.42%, profit_factor 5.11, real BUY / SELL 90/90
• delta: avg_return 0.56%, profit_factor 1.06, BUY overlap -11, SELL overlap -11

Risk And Quality

• avg MFE / MAE: current 6.93% / -2.57% vs refined 7.62% / -2.57%
• short 00-05d: current -1.72% vs refined -1.67%
• short 06-10d: current -0.69% vs refined -0.59%

Walk-Forward

• anchored expanding: current 8/9, avg 4.00% vs refined 9/9, avg 4.52%
• rolling 3Y: current 7/7, avg 5.19% vs refined 7/7, avg 5.76%

Removed-Trade Attribution

• removed trades vs current alpha: 11
• removed-set avg_return / win_rate: -1.81% / 0.00%
• recommendation mix KEEP / OBSERVE / OVER: 11/0/0
• Interpretation: the refined branch now improves by removing only weak, losing short-holding glued trades; it does not rely on deleting profitable samples.

Upgrade Gates

• headline_quality_gate: requires avg_return delta >= +0.30%, profit_factor delta >= +0.50, and no short-bucket deterioration. result: PASS
• stability_gate: requires anchored and rolling walk-forward to be no worse than the current formal alpha branch. result: PASS
• removal_quality_gate: requires OVER_REMOVAL = 0, OBSERVE_REMOVAL <= 1, removed-set win_rate <= 5%, and removed-set avg_return < 0. result: PASS
• alignment_cost_gate: automatic promotion only if incremental overlap loss is no more than 8 additional BUYs and 8 additional SELLs versus the current formal alpha branch. result: FAIL

Final Decision

• governance_decision: DUAL_TRACK_GOVERNANCE
• Decision: keep alpha_first_selective_veto as the formal alpha branch and keep alpha_first_glued_refined_hot_cap as the governed leading candidate.
• Reason: quality, stability, and removal quality all pass, but alignment loss is still large enough that promotion should be explicit rather than automatic.

Recommendation

• Keep alpha_first_selective_veto as the current formal alpha branch for now.
• Keep alpha_first_glued_refined_hot_cap as the first governed promotion candidate if the governance objective explicitly shifts toward stronger alpha.