Struct AttentionState 

pub struct AttentionState {
    pub learned_weights: [f32; 7],
    pub history: Vec<AttentionOutcome>,
    pub last_recall_ms: u64,
    pub total_recalls: u32,
}

Fields

learned_weights: [f32; 7]

Learned optimal weights (running average from good outcomes).

history: Vec<AttentionOutcome>

Recent outcome history.

last_recall_ms: u64

Last recall timestamp for quality inference.

total_recalls: u32

Total recalls tracked.

Implementations

impl AttentionState

pub fn load_or_init(output_dir: &Path) -> Self

pub fn compute_attention(&self, signals: &AttentionSignals) -> AttentionVector

Compute attention vector from context signals.

pub fn infer_quality(&self) -> f32

Infer quality of last recall from time gap. Returns quality score (0.0 = bad, 1.0 = good).

pub fn record_outcome(&mut self, quality: f32, weights: &[f32; 7])

Record the outcome of a recall (the quality applies to the PREVIOUS recall).

pub fn mark_recall(&mut self)

Mark that a recall just happened (for next quality inference).

pub fn save(&self, output_dir: &Path) -> Result<(), String>

Save to binary.