Struct PredictiveCache 

pub struct PredictiveCache {
    pub predictions: Vec<Prediction>,
    pub stats: CacheStats,
}

Fields

predictions: Vec<Prediction>

stats: CacheStats

Implementations

impl PredictiveCache

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

pub fn check(&self, query_hash: u64) -> Option<(Vec<u32>, f32)>

Check if we have a prediction for this query. Returns cached blocks + confidence. This is called BEFORE the actual search, so the results can be used immediately.

pub fn evaluate( &mut self, query_hash: u64, actual_results: &[u32], thought_graph: &mut ThoughtGraphState, ) -> (&'static str, usize)

Evaluate prediction accuracy after a recall completes. Compares predicted blocks against actual results. Returns: (hit_type, overlap_count) where hit_type is “hit”, “partial”, or “miss”.

pub fn predict_next(&mut self, thought_graph: &ThoughtGraphState)

Generate predictions for the next likely query based on current session state. Called after each recall to pre-load the cache.

pub fn export_stats(&self) -> (u32, u32, u32, f32)

Export stats for cross-instance exchange.

pub fn merge_stats( &mut self, remote_predictions: u32, remote_hits: u32, remote_misses: u32, )

Merge remote stats (additive, for reporting).

pub fn dream_cleanup(&mut self)

Dream cleanup: remove predictions with very low confidence.

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

Save to binary.