Agent Experiment provides a Java lifecycle for repeatable AI-agent evaluations: load versioned dataset items, provision isolated workspaces, invoke an agent, evaluate outcomes, and persist auditable results for comparison and re-evaluation.

Canonical documentation: lab.pollack.ai/projects/agent-experiment

Agent Experiment 0.7.0 requires Java 21. Add the modules your experiment needs; all three published modules use the same version.

<dependency>
    <groupId>io.github.markpollack</groupId>
    <artifactId>experiment-core</artifactId>
    <version>0.7.0</version>
</dependency>

From a clean checkout, run:

./mvnw clean verify

Agent Experiment is a released, pre-1.0 framework, so APIs may change between minor versions. Version 0.7 uses Agent Judge 0.15 while retaining tested readers for Agent Experiment 0.5 / Judge 0.13 result files. Run records support audit and comparison; exact replay still requires pinning external models, agent CLIs, tools, and services.

Business Source License 1.1