feat(test_case): make trace_dict public for post-hoc agentic evaluation

[tiffanychum]

Summary

• LLMTestCase._trace_dict → public trace_dict field — the private attribute made it impossible to pass a pre-recorded trace at construction time, so the four agentic trace metrics (TaskCompletionMetric, StepEfficiencyMetric, PlanQualityMetric, PlanAdherenceMetric) only worked with @observe at runtime.
• Alias support — serialization_alias="traceDict" + validation_alias=AliasChoices("traceDict", "trace_dict") so both snake_case and camelCase work in model_validate and JSON round-trips (consistent with all other fields on LLMTestCase).
• All internal usages updated — 6 assignment sites in evaluate/execute.py and all 4 metric files updated from ._trace_dict to .trace_dict; the runtime @observe path is unchanged.

Motivation

The non-trace evaluation path in task_completion.py is already marked:

# TODO: Deprecate this soon

But until now there was no way to reach the trace path without @observe at runtime. This PR closes that gap, enabling:

• Offline / batch evaluation from saved logs or databases
• CI trace replay without re-running the agent
• Third-party pipelines where you can't decorate the agent code
• Post-mortem analysis of production traces

Changes

File	What changed
deepeval/test_case/llm_test_case.py	_trace_dict: PrivateAttr → trace_dict: Field(...) with aliases
deepeval/evaluate/execute.py	6 internal assignments updated to public field
deepeval/metrics/task_completion/task_completion.py	_trace_dict → trace_dict (4 references)
deepeval/metrics/step_efficiency/step_efficiency.py	_trace_dict → trace_dict (6 references)
deepeval/metrics/plan_quality/plan_quality.py	_trace_dict → trace_dict (4 references)
deepeval/metrics/plan_adherence/plan_adherence.py	_trace_dict → trace_dict (8 references)
tests/test_core/test_test_case/test_single_turn.py	Updated assertions + new test_trace_dict_constructor_and_alias test
examples/tracing/test_posthoc_evaluation.py	New example showing post-hoc evaluation from a saved trace

Test plan

• python -m pytest tests/test_core/test_test_case/test_single_turn.py -v — all existing tests pass, new alias test passes
• python examples/tracing/test_posthoc_evaluation.py — post-hoc evaluation runs end-to-end with a pre-recorded trace dict
• Existing @observe runtime path: no behaviour change (assignments in execute.py write to the same field)

Related

Parallel to issue #2579 (Turn._mcp_interaction PrivateAttr bug) — same root cause: Pydantic v2 PrivateAttr fields can't be set via the constructor.

[vercel]

@tiffanychum is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.