feature/PAAL-212-trace-ids-in-testworkflow

[qa-jil-kamerling]

OTEL Tracing Implementation

Phase 2: test run

• Created otel_setup.py module and instrumented run.py to wrap each agent query in a span with trace context propagation via HTTPX, enabling end-to-end trace visualization in Tempo/Grafana
• Captured trace IDs in experiment results

Phase 3: evaluation

• Each test case in ragas_experiment.jsonl now includes a trace_id field linking evaluation results to distributed traces,
  enabling correlation between test execution and agent behavior
• Fixed trace ID preservation through evaluation
• Modified evaluate.py to manually preserve trace IDs when RAGAS processes experiment data (RAGAS drops custom fields), ensuring evaluation_scores.json maintains trace context for debugging failed tests

---

Dependencies added: opentelemetry-api, opentelemetry-sdk, opentelemetry-exporter-otlp-proto-http, opentelemetry-instrumentation-httpx

[felixk101]

ggf. magst du einfach @README.md verwenden

[qa-jil-kamerling]

Danke, aber wir hatten uns entschieden Claude.mds und readmes zu trennen

[felixk101]

Ich hatte mit dem Code schonmal Probleme. Es wird manchmal NaN als Zahl in die JSON-Outputdatei geschrieben. Das ist allerdings nicht valides JSON. 😕

In meinem Code musste ich in publish.py folgendes einbauen, was nicht so schön ist.

def _is_metric_value(value: Any) -> TypeGuard[int | float]:
    """Check if a value is a valid metric score (numeric and not NaN)."""
    if not isinstance(value, (int, float)):
        return False
    if isinstance(value, float) and math.isnan(value):
        return False
    return True

(Das ist aber nicht Teil des Tickets)

FYI @fmallmann