Check the answer before you use it.
A practical way to separate a fluent answer from one you can support.
Begin with the claim that matters.
A fluent answer can make several kinds of statement at once: a fact, an estimate, a recommendation and an assumption. Before using the answer, identify which part your decision depends on.
Copy that claim into a separate note. Make it specific enough to check. “This tool is better” is too broad. “This tool exported the required format in this run” is a claim you can inspect.
Find the underlying evidence.
If the answer cites a source, open it. Check that the source exists, that it says what the answer claims, and that its date and scope fit your question. A source about a different version or situation may be real and still fail to support the claim.
- Prefer the original document, test output or source record.
- Separate what the evidence shows from what the answer infers.
- Mark missing evidence as missing. Do not fill the gap with confidence.
Test the part you can test.
For a workflow or software task, choose one small example with a result you can recognize. Save the input, output, tool version and date. Inspect the actual artifact, not just the tool’s success message.
For example, if an assistant says it removed identifying details from a document, inspect the delivered document for those details. A successful export proves an export happened. It does not prove the content is right.
Record the limit.
Write down what passed, what failed and what remains uncertain. One successful example supports a claim about that example. It does not establish that every future case will work.
Use this note: Claim / source or test / observed result / remaining uncertainty / next action.
Make the next move proportional.
A low-impact draft may need a quick check. A decision affecting money, access or another person deserves stronger evidence and the right human review. Set the check to the consequence.
An introductory TQO guide. This is a suggested practice, not a report of a test Tee has personally run.
All playbooks