Wanaku

Appearance

Test Wanaku Semantic Router ​

This page describes what the tests check, how to evaluate classification quality, and the test suites that need more than a Maven build.

Validation scope ​

The reference checks cover native semantic selection, auxiliary Kamelet loading, MCP discovery and invocation, explicit no-match, provider failure, malformed results, archive safety, digest checks, runtime compatibility, fetch failure, failed Camel startup, failed registration cleanup, accepted POST response loss, full HTTP completion deadlines, action failure, preview timeout and capacity, registration ordering, shutdown, and a packaged JAR smoke test. Deterministic fixtures check integration behavior. They do not measure classification quality.

Evaluate classification quality ​

For model evaluation, use a separate deployment and a representative set of saved examples. Run semantic evaluation previews first. Compare each expected label with its actual label. Review no-match and error cases separately. Use demonstration actions before enabling operations with backend side effects. Record the provider, model, revision, and distribution or image digest with the results.

Wolf Defender smoke test ​

Run the provisioned-model smoke test with WOLF_MODEL_DIRECTORY=/models/wolf-defender mvn -Dtest=WolfDefenderIntegrationTest test. Without that variable the test is skipped; deterministic guard routing tests cover rejection, evaluation failure, and accepted dispatch without inference.

Initial verification used a locally installed Camel 4.23.0-SNAPSHOT Wolf Defender JAR (Maven metadata timestamp 20261009105227, SHA-256 c69acd50a02108d28292b733bd334bdeaaff399c32669df60cad2aadf30a5da0). Camel PR 27564 was merged as 5073253e66bdfa0f9f8cc80c8900a97466b0740c. These identify the verified artifact and upstream change, rather than pinning future snapshot resolution. Release builds must verify that the selected Camel snapshot contains the native injection operation and modelDirectory property.

Next steps ​