Retrieve Experiment Span

Retrieve single experiment trace with full span tree, or submit workflow result.

GET /evaluations/experiments/<experiment_id>/logs/<trace_unique_id>/ Returns:

• Trace-level aggregated metrics (cost, tokens, duration, etc.)
• Full hierarchical span tree with all children
• Enriched with storage (input/output) for API key authentication
• Complete scores (LLM evaluator + human annotation) with evaluator metadata

PATCH /evaluations/experiments/<experiment_id>/logs/<trace_unique_id>/ Body: {"input": {...}, "output": {...}, "metrics": {...}, "metadata": {...}} Purpose: Submit wait-task workflow result via resume mechanism

• Uses get_full_object_by_unique_id for cached retrieval (performance optimized)
• Supports both JWT and API key authentication
• Partial updates with existing data merging
• Creates workflow spans in unified format with proper trace hierarchy
• Triggers evaluators if specified in experiment configuration
• Input/output can be any JSON type (dict, list, string, number, boolean)

Note: The URL parameter is called 'log_id' but it should be the trace_unique_id. This returns a full TRACE (aggregated) with span tree, not a single log/span. The 'id' field in the list endpoint exposes trace_unique_id for use in detail/PATCH operations.

Uses SpanTreeSerializerContextMixin to automatically handle:

• Span tree inclusion (always enabled for detail view)
• Storage enrichment based on authentication type (JWT vs API key)
• Dynamic serializer configuration

Uses DataEnrichmentMixin to enrich scores with:

• Human annotation scores from Postgres EvalResult
• Evaluator metadata (name, slug, score_value_type)
• All score types (numerical, boolean, string, categorical, json)