Module case 

The evaluation case format — JSON trace fixtures for deterministic replay.

Structs

LlmTrace

A complete LLM conversation trace loaded from a JSON fixture.

TraceExpects

Declarative expectations for grading a run.

TraceStep

A single LLM response step within a turn.

TraceToolCall

A tool call within a trace response.

TraceTurn

A single conversation turn (user input + scripted LLM response steps).

Enums

TraceResponse

The response content for one step — either plain text or tool calls.

Functions

load_suite

Load every *.json trace fixture in dir, sorted by path for stable ordering.