ADR-001: OpenTelemetry/OpenInference as telemetry foundation

Decision

Use OpenTelemetry as the telemetry transport/data-model foundation and OpenInference/GenAI semantic conventions for AI-specific attributes where applicable.

Context

A proprietary trace protocol would reduce interoperability and create maintenance cost.

OpenTelemetry provides stable tracing APIs/SDKs and semantic conventions. OpenInference adds AI-specific concepts such as LLM, embedding, chain, tool and retrieval information.

Consequences

Positive: - ecosystem compatibility - easier instrumentation - standards-oriented portfolio signal - future interoperability

Negative: - semantic conventions evolve - normalization layer required - not all AI providers expose identical information

Implementation

Create:

backend/src/opengrade_server/tracing/
  ingestion.py
  normalization.py
  otel.py
  openinference.py
  redaction.py

Keep vendor-specific normalization outside domain entities.