Pareto
Candidate pool with Pareto-inspired selection for GEPA's evolutionary search.
All candidates are kept (never pruned). The Pareto property is enforced implicitly through selectCandidate: selection is weighted by how many validation instances each candidate is the best performer on ("coverage"). Dominated candidates (leading on zero instances) have zero selection probability and are effectively ignored.
This avoids explicit pruning while still biasing selection toward the Pareto frontier. getBestCandidate returns the candidate with the highest aggregate score regardless of coverage.
Functions
Adds candidate to the pool, updating per-instance best-score bookkeeping.
Snapshot of all candidates in insertion order.
Returns the candidate with the highest ParetoCandidate.aggregateScore, ignoring coverage.
Selects a candidate probabilistically from the Pareto frontier.