Ui UdZddlZddlmZddddddd d dd d dd Zeeeeeffed< ddedededefdZ ddededededef dZ ddedededededef dZ y)alEbbinghaus forgetting curve mathematics. R = e^(-t/S) Where: R = retention (0.0 to 1.0) t = time since last access (hours) S = memory strength = base_strength * access_multiplier * surprise_bonus The Ebbinghaus forgetting curve models how memory retention decays exponentially over time. This implementation extends the basic model with two reinforcement factors: 1. Access reinforcement (spaced repetition): Each access strengthens the memory using logarithmic scaling -- the first few accesses have the biggest impact, diminishing returns after that. 2. Surprise bonus: Higher surprise scores produce more vivid memories that resist decay longer. Reference values (base_strength=24.0, access_count=1, surprise_score=0.5): - t=0h -> R=1.0 (just accessed) - t=24h -> R~0.37 (e^-1) - t=48h -> R~0.14 (e^-2) - t=72h -> R~0.05 (e^-3) N)Dictg?)threshold_daysretention_floorg333333?Zg?) aggressivemoderate conservativeinfiniteDECAY_POLICIES access_countsurprise_score base_strengthreturnctd|}tdtd|}td|}dtj|z}dd|zz}||z|z}|S)aCalculate memory strength S from access count and surprise score. S = base_strength * access_multiplier * surprise_bonus Access multiplier uses logarithmic scaling (not linear) so that the first few accesses have the largest effect: multiplier = 1 + ln(access_count) Surprise bonus scales linearly from 0.5 (score=0) to 2.0 (score=1): bonus = 0.5 + 1.5 * surprise_score Args: access_count: Number of times this memory has been accessed. Clamped to minimum 1. surprise_score: Surprise score in [0.0, 1.0]. Clamped. base_strength: Base strength in hours. Default 24.0 means ~37% retention after 24h with no reinforcement. Returns: Memory strength S (positive float, in hours). r ??g?)maxminmathlog)rrraccess_multipliersurprise_bonusstrengths ./mnt/e/genesis-system/core/rlm/decay_curves.pycalculate_strengthr +sk6q,'Lc#~67N]+Mdhh|443//N00>AH Ohours_since_accessc|dkryt|||}tj| |z }tdt d|S)aFCalculate memory retention using Ebbinghaus forgetting curve. R = e^(-t/S) Where: t = hours_since_access S = calculate_strength(access_count, surprise_score, base_strength) Args: hours_since_access: Time since memory was last accessed, in hours. Non-negative. Values <= 0 return 1.0. access_count: Number of times this memory has been accessed. surprise_score: Surprise score in [0.0, 1.0]. base_strength: Base strength in hours. Returns: Retention value clamped to [0.0, 1.0]. rrr )r rexprr)r"rrrr retentions rcalculate_retentionr&TsL2Q!, NH,,x78I sCY' ((r!policyc tj|}|,td|dttj |ddkry|ddz}|dkry||kryt ||||}|d}||kr |dkr|d k\ry y |d kry y) aDetermine decay action for a memory given its state and policy. Args: hours_since_access: Time since last access in hours. access_count: Number of times accessed. surprise_score: Surprise score in [0.0, 1.0]. policy: One of "aggressive", "moderate", "conservative", "infinite". base_strength: Base strength in hours. Returns: One of: "retain", "demote", "delete" zUnknown decay policy: z . Valid: rrretain8@rg?demotedeleter)rget ValueErrorlistkeysr&) r"rrr'rconfigthreshold_hoursr%rs r should_decayr4ys&   'F ~1&<##'(;(;(=#>"?AB B!#-.5OD O+#L.-I./O?"|a'7 T !3 r!)r*)rrr*)r r*) __doc__rtypingrrstrfloat__annotations__intr r&r4r!rr<s4 &'3?#%#>')cB#%#> /S$sEz**+ &&&& &V ")")")") ")  ")R 5555  5  5  5r!