Rubric
A ‘rubric’ is a structured list of criteria or questions that the LLM-as-a-judge evaluator uses to score the LLM’s response.Span
A ‘span’ records a specific operation (like an LLM request, tool call, etc.) within a trace. A span contains details about the operation, performance metrics, metadata, input/output of the operation, and links to parent and child spans.NOTE: In Adaline, a prompt represents a type of LLM request happening in your AI agent / workflow. Each span sent to Adaline capturing a LLM request operation should ideally be associated with a prompt.
Test Case
In Adaline, a ‘test case’ is a set of variables values.Trace
A ‘trace’ maps the entire lifecycle of a request served by your AI agent / workflow. A trace contains details about the request, performance metrics, metadata, and one or more spans.NOTE: In Adaline, a project represents an AI agent / workflow and each trace sent to Adaline must be associated with a project.