Deployment Policy A/B Pre-registration

Scope

Research-only initialization replay. Production v1.3, P1, and P2 are unchanged.

Policies

- fresh_gradual: empty account; maximum two new v1.3 signal entries at each eligible weekly snapshot.

- legacy_copy_full: on launch day, buy the observable P1 v1.3 holdings at their current weights; afterwards use the same signal-entry rule.

- legacy_copy_staged4: buy one quarter of the observable launch-day P1 target weights on each of four weekly snapshots; afterwards use the same signal-entry rule.

Copied initialization is an explicit onboarding-policy exception: it copies existing weights without reapplying entry-time industry/chain buy caps. New subsequent signal entries apply current v1.3 entry caps.

Eligible Launch Dates

All weekly v1.3 snapshot dates in the formal sample where M-filter is ON and an observable P1 v1.3 position snapshot exists. This prevents copying positions when the model would prohibit deployment.

Outcomes

Forward account return, maximum drawdown, average exposure, and ending holdings at 1m/3m/6m/12m. Primary policy judgment uses paired differences versus fresh_gradual.

Timing Bias Control

Weekly cohorts overlap; raw cohort counts are descriptive only. Uncertainty is summarized by bootstrapping launch-year blocks, and results are separately shown for 2015-2017, 2018-2023, and 2024-2026.

Decision Rule

Do not recommend changing deployment policy unless an alternative exceeds fresh_gradual in paired median return at both 3m and 12m, does not worsen median MDD by more than 2 percentage points at either horizon, and the 90% year-block bootstrap interval for return delta is above zero at both horizons. Otherwise retain current gradual deployment default.

Limitation

This isolates account initialization and ordinary exits/new entries. It omits smart-capacity replacement after a copied portfolio because copied onboarding positions do not possess a comparable original entry-score state; a positive result would require a second implementation-sensitivity test before any production action.