Stress-test a data analysis before you present it
Find the alternative explanation your chart doesn't rule out before someone else does.
- Use it for
- Anyone about to present numbers that will drive a real decision.
An analysis that supports the conclusion you were already leaning toward is the easiest kind to produce and the most dangerous kind to trust, because you stop looking the moment the numbers agree with you. The question that actually protects you isn't "does this chart show what I think it shows" — it's "what else could produce this exact pattern, and does my data rule that out." Almost nobody asks that question of their own analysis before presenting it, and it's the first question a skeptical audience will ask for you, at the worst possible time.
This prompt exists to ask it first, while there's still time to check.
When not to use this
Needs a real analysis with actual numbers and a stated conclusion — "is my data good" with no specifics produces nothing to stress-test. And it won't catch errors in the underlying data itself, only in the reasoning from data to conclusion; a corrupted dataset needs a different kind of check entirely.
Did this work?
You can name the strongest alternative explanation for your finding and say plainly whether your data rules it out — not just restate the finding more confidently.
Tested on claude-opus-5. Evidence status is draft; it moves to battle-tested only on recorded runs, never by hand.