DjXdZddlmZddlZedGddZdddd d Zy) a?Experimental state-aware action replay; disabled in the production strategy. Actions are decisions at the close, not broker fill states. A deferred score exit is latched until the minimum holding age, so a one-bar crossunder is not lost. Trailing stops always override holding. Entry blocks never queue stale entries. ) dataclassNT)frozenc>eZdZUdZeed<dZeed<dZeed<dZy) StatePolicyr macro_memory exit_cooldownmin_holdc|j|j|jfD]2}t|trt|t s|dks)t dy)Nrz1Policy durations must be nonnegative integer bars)rrr isinstanceintbool ValueError)selfvalues F/Users/jameslopez/projects/TradingBot25/strategies/mlp_state_policy.py __post_init__zStatePolicy.__post_init__sJ''););T]]KEeS)Zt-DPQ  !TUULN) __name__ __module__ __qualname__rr __annotations__rr rrrrr s&L#M3HcVrrF)risktime_ok dead_zonec t|| tjtntj| t} | tj tntj| t} t fd|||||| | fDr tdtjt} tjttjt}} tjdd}tjttjt}}d\}}}d\}}}d}tD]}| |r||jz}||kxs||k}|r||d zk7rt|||}|d kDxr|||d |z zk}|r||n||}| |r|xs t|}||z |jk}|xr|xr | xr| |||<| |r&|s|r"|s d\}}}d ||<d ||<||jz}nMd ||<nGt||xr |xr| |||<||r&|s$| |rd }d | |<d||<|||}}| xr|||k}|| |<| | ||||dS)a/Replay raw opportunities using position age, exit memory and macro risk. A risk bar blocks itself and N following bars, refreshed on every risk bar. An exit blocks the following N bars; entry becomes possible at exit+N+1. Risk is entry-only: it does not liquidate an existing position. )dtypec3dK|]'}tj|jfk7)yw)N)npasarrayshape).0ans r z&replay_state_policy..$s) j0i12::a=  1$ &0is-0z7Replay arrays must be one-dimensional and equally sized stay_flatzrWsC " $VVV=A$"'4.r