Expand description
Retrieval Objective implementations for khive-runtime.
Domain-specific objectives that operate on pre-computed retrieval signals. Pure math: no IO, no async. The runtime layer materialises the signal data and feeds it in via the candidate struct.
See ADR-061 — Retrieval Infrastructure. See ADR-033 — Recall Pipeline (NoteCandidate, DecayAwareImportanceObjective, TemporalRecencyObjective, RerankerObjective).
Structs§
- Decay
Aware Importance Objective - Scores a
NoteCandidateby salience with configurable temporal decay. - Graph
Proximity Objective - Scores a candidate by graph proximity to anchor nodes.
- Note
Candidate - Pre-computed signals for a single memory note candidate.
- Reranker
Objective - Scores a
NoteCandidateusing a named reranker’s pre-computed score. - Retrieval
Candidate - Pre-computed retrieval signals for a single candidate entity.
- RrfFusion
Objective - Scores a candidate by its pre-computed RRF fusion score.
- Temporal
Recency Objective - Scores a
NoteCandidateby pure temporal recency with a configurable half-life. - Text
Relevance Objective - Scores a candidate by BM25/FTS relevance.
- Vector
Similarity Objective - Scores a candidate by cosine similarity to the query vector.