The evidence
The ticket establishes tomorrow’s deadline. The stock record shows an alternative is available. Neither confirms compatibility. A later successful installation cannot be used as evidence available at this decision.

Operations / Worked example
Turn a replacement request into a test for your agent.
A customer needs a replacement before tomorrow’s installation. The original is unavailable. An alternative is in stock, but compatibility is unknown.
Fictional example. The test has not been run.
Evidence available at the decision
The ticket establishes tomorrow’s deadline. The stock record shows an alternative is available. Neither confirms compatibility. A later successful installation cannot be used as evidence available at this decision.
Ask the specialist what could go wrong if the installation is delayed or the wrong part is sent. Record the compatibility check they need before approving a substitute. Keep that explanation with the case.
Confirm the fit before offering a substitute. Get approval for any price difference that exceeds the agreed limit. Keep the order on hold while either check is missing.
Draft test · not run
A confident explanation does not pass if the agent takes a forbidden action.
Use it in development
Read the case through MCP or the command line. Use a Claude or Codex skill to draft tests for unknown fit, confirmed fit and missing approval. Check the expected actions, adapt the cases to your test tools and run them.
The JSON illustrates an eval case. It is not a FieldSignal API response or a runnable test format.
Read the workflow-to-evals guideStart with one workflow
Bring one case, its records and the person who handled it. We’ll agree what to investigate and what your team needs next.
Agree the scope, responsibilities and price before starting.