Plan Adherence
Plan Adherence is a single-turn metric to determine if your agent followed the plan it laid out
Overview
The plan adherence metric is a single-turn metric that uses LLM-as-a-judge to assess whether your LLM agent actually followed the plan it laid out for completing a task.
How Is It Calculated?
The plan adherence metric uses an LLM to extract the task and the plan from your agent's reasoning steps, then uses the same LLM to score how closely the execution steps in the trace follow that plan.
The final score is the alignment of the execution steps with the task and plan as extracted from the trace.
Usage
To run the plan adherence metric on Confident AI, add it to a single-turn metric collection. This will allow you to use plan adherence metric for:
- Single-turn E2E testing
- Online and offline evals for traces
Last updated on