Module relevance 

Shared lexical-spatial ranking for every natural-language recall surface.

The legacy ranker subtracted a fixed keyword boost from squared spatial distance and clamped the result at zero. On real memories this collapsed many distinct candidates to the same score, making traversal order decide the final ranking. This module keeps the score continuous and normalizes lexical evidence by query coverage.

Structs

RelevanceQuery

Parsed query reused while scoring all candidate memories.

Functions

apply_boost

Apply a positive cognitive boost without collapsing distinct ranks to zero.

apply_confidence_boost

Apply an evidence-confidence boost to the rank distance.

merge_candidates

Build the block set that recall ranks, from the two independent candidate sources.

rank_distance_from_score

Lower-is-better rank distance for a lexical score already computed during candidate filtering. Keeping this separate prevents normalizing every candidate twice on the hot recall path.