01
Describe performance for one visible task
The fictional task is to find the stated meeting date in a short approved document and return the passage containing it. The review record would contain the question, document version, proposed answer and supporting passage. A reviewer could compare the answer with the text and record an unresolved case when the document contains conflicting dates. This defines what would be examined without inventing a success rate. It also prevents fluent wording from standing in for evidence that the answer matches the selected document.
02
Ask separately where the task extends
Now consider a second document with a scanned page and a third written in another language. Whether the same workflow handles those inputs is a separate question from its performance on the first document. A brief can list those cases as unexamined rather than imply that one example establishes broad competence. The worksheet uses a coverage column for each input type and leaves unsupported cells open. This makes the boundary visible while avoiding a claim that a small document exercise measures general intelligence or a future research horizon.
03
Specify authority independently of the answer
The fictional assistant may propose a date and show its source, but it may not send invitations or edit the calendar. A human decides what to do with the suggestion. That authority boundary is independent of whether the proposed answer is correct. A more capable answerer need not receive more permission automatically. The completed brief records the task, the unexamined input types and the allowed actions in separate fields. This gives Witasi a practical company canvas while preserving the distinction between a naming association and a demonstrated capability.