Module bayesian_scorer

Bayesian retrieval scoring for multi-factor candidate ranking (RFC-013 Part C).

Replaces flat cosine scoring with a weighted composite:

score = w_sim * similarity
      + w_recency * recency_factor
      + w_reward * reward
      + w_success * success_score
      + w_region * region_match

Weights are configurable and can be learned from online feedback (logistic regression on outcome data).

Structs

CandidateFeatures

Features for a single retrieval candidate.

ScoringWeights

Scoring weights for Bayesian retrieval ranking.

WeightLearner

Online weight learning from outcome feedback.

Functions

rerank

Re-rank a list of candidates using Bayesian scoring.

score_candidate

Compute the Bayesian composite score for a candidate.