Struct DistributionalRefiner 

pub struct DistributionalRefiner;

Refines atom vectors by blending them toward distributional context vectors.

For each atom A that participates in relations {R1, R2, …}:

1. For each relation Ri, collect all peer atoms that also participate in Ri
2. Build a context vector by frequency-counting indices across peer atoms
3. Blend: refined_A = merge(original_A, context_vector, alpha)

The merge preserves sparsity by taking the top-k indices from the weighted union of original and context indices.

Implementations

impl DistributionalRefiner

pub fn refine( atom_memory: &AtomMemory, triple_store: &TripleStore, config: &RefinerConfig, ) -> RefinementReport

Run one refinement pass over all atoms.

This modifies atom vectors in-place. Each atom’s vector is blended toward the centroid of its relational peers. Requires write access to AtomMemory (call via Governor, not from the read-only cognition loop).