1. Name the options
“What is the best search tool?” has no useful boundary. “For our two-person team with three weeks to launch, should we build search or use a hosted service?” gives the model an actual choice. Enter each option separately so you can inspect the whole spread.
2. Give the facts that matter
Include your goal, budget, deadline, team capacity and the cost of a wrong call. Leave out unrelated detail. Jev weighs the situation you enter; it does not gather evidence on its own. Write one fact per sentence. A stress test checks each sentence separately, so this makes it more useful.
A question template
“Given our situation, which of A, B or C best supports our goal?” Then list the concrete constraints, such as time, money and people, one sentence each in the situation field.
3. Pick a format
Use Compare when the answer is one of a few paths. Use Yes or no for a single claim or action. Use Score for an ordered rubric such as priority or urgency. Don’t treat a score as a precise measurement between the levels.
4. Treat a close result as a reason to find more facts
If the options are close, the context check is weak, or your result says “Too close to call,” ask what information would separate them. A single price quote, capacity estimate or user interview may be worth more than another model call.
5. Find the fact the decision depends on
Run a stress test when the stakes are real. If leaving out one fact flips the call, that fact carries the decision. Verify it before anything else.
6. Check changing facts yourself
Jev’s read has no citations. Prices, product limits, laws and policies change. Read the current primary source before a costly or hard-to-reverse choice. A high probability is not proof that a premise is true.