This guide describes the Agent Eval 0.18.0 API. Maven coordinates and Java packages still use
agent-judge and io.github.markpollack.judge.Judge.judge() and Jury.vote() take no execution argument. Construction binds the actual requirement/roster, evidence and lower runtime.
EvalRuntime.execute(request) returns the native answer and its Invocation; EvalModel.generate(request) supplies a generated answer.
A Verdict retains collective and individual judgments, seats, children/attempts, roster or explicit parent, provenance, and invocation owners.
conclusion() derives PASS, FAIL, INCONCLUSIVE, or NOT_APPLICABLE. requireUsable() validates and returns the unchanged object.
Core owns no Jackson engine. Add the supported Jackson 2 module explicitly; use NativeRequirementCodecs.codec() for RFC/EARS values and VerdictCodec for text/AllOf.
Descriptions stay V3; results/evaluations use V6.
Source at v0.18.0: producer README, writing judges, describing juries, V6 storage.
These pinned sources are distinct from released 0.17 Javadoc.
For source and stored-result changes, follow Migrating to 0.18.