OpenTelemetry Semantic Conventions Conformance

Does an instrumentation actually emit what the semantic conventions say it should? This repo answers that the same way for every library, every implementation and every language: run a small program that exercises the library, collect what it emits through Weaver live-check, and check it against expectations declared in YAML.

tools/runner/	the runner. Generic — it carries no semantic conventions of its own
tools/gen-ai/	what makes a run a GenAI run: the registry pin, the advice policies, and a mock LLM server so scenarios are deterministic without cassettes
tools/http/	the same for HTTP: the upstream registry pin, and the test client that drives both sides of the domain
tools/java/	what every JVM scenario shares, in any domain: the launcher that builds and runs one, the Gradle convention plugins, and the SDK bootstrap
tools/js/	what every Node scenario shares: the launcher that installs a build, and the SDK bootstrap
scenarios/gen-ai/	the GenAI scenarios and the coverage they produce
scenarios/http/	the HTTP scenarios and the coverage they produce

A conformance directory names the wrapper it wants under runner:, so one command runs any of them:

pip install -e tools/runner -e tools/gen-ai/mock-server -e tools/gen-ai/runner
otel-conformance path/to/directory --report-only

See the runner's README for what a scenario and its conformance.yaml look like.

Maintainers

• Christophe Kamphaus, Independent
• Jay DeLuca, Grafana Labs
• Josh Suereth, Google
• Liudmila Molkova, Google
• Trask Stalnaker, Microsoft

For more information about the maintainer role, see the community repository.

Approvers

• None

For more information about the approver role, see the community repository.