Skip to content

task

Diagnose what needs to change

Choose an intervention after checking whether the failure comes from requirements, representation, available context, or implementation.

Diagnose what needs to change

Choose an intervention after checking whether the failure comes from requirements, representation, available context, or implementation.

Prerequisites

An interpretable Case result, its applicable standards, input and materials, and the candidate identity.

Steps

  1. Read the actual output and criterion. Confirm the expected behavior is a requirement your experts support.
  2. Check whether the Case supplies the source information and conditions needed to meet that requirement.
  3. Inspect whether the benchmark represents the relevant behavior slice and whether the failure repeats across comparable Cases.
  4. Direct the next assignment to the owner: an expert question, a coverage or Case update, a Lemon Code intervention, a Lemon Goal experiment, or targeted training material.

Object and state changes

Diagnosis itself does not change the standard or candidate. Each accepted follow-up uses its owning workflow.

Success criteria

The next assignment states a measurable behavioral problem, relevant Cases, and the reason the proposed intervention addresses it.

Common failure modes

Changing prompts before checking missing context; treating every rubric failure as a model-weights problem.

Found something unclear?

Report outdated, unsupported, or confusing docs so we can fix the source page.

Report a docs issue