v0.5.0

0.5.0 (2026-10-16)

Features

• bench: build the release evidence report, and the lint its metric is (#311) (f862513)
• bench: measure the router against the best single method it must beat (#307) (9ee2c99)
• bench: measure what a warm start saves, in evaluations (#312) (d88a5ed)
• bench: measure what compiler-guided repair recovers, and what it costs (#305) (4756182)
• bench: measure what novelty detection and experience reuse suppress (#306) (f5a4d12)
• bench: tune a consumer model with the engine against grid search at equal budget (#310) (dfddeaf)
• eng: diff openevolve's config against the catalog, not against prose (#315) (5097fef)

Refactoring

• core: handle nulls in the three worst offenders instead of suppressing them (#313) (5109363)
• core: handle the pipeline's nulls instead of suppressing them (#304) (7bad10a)