The Stress Test is a Confidence tool. It reads your prompt as a whole — role, context, task, constraints, output — and names the tensions that would make a real model drift, hedge, or invent. You see the issues before your reader does.
Inside the Practice Lab. Click Run Stress Test on any prompt.
What it checks for
Conflicting blocks
A persona that says "be concise" alongside an output spec that demands a 2,000-word memo.
Missing context
A task that assumes information the prompt never supplies.
Ambiguous instructions
Phrases that two readers would interpret differently — likely the AI will too.
Unverifiable claims
Asks the model to assert facts it has no way to check. Flags so you can require sources.
How to use the findings
Read findings in order of severity
The top finding usually unblocks several others when fixed.
Edit the prompt, not the AI reply
Stress Test feedback is about the brief, not the answer. Tighten the brief.
Re-run
After each meaningful edit. Watch the score and findings move together.
Save the strong version
When the test is happy, save back to your Library so this prompt starts strong next time.
One pass of the Stress Test on the prompts you run weekly compounds for months. It's cheap insurance against confident-but-wrong output reaching a client.