Pattern 04 — Evidence before confidence

Thesis

Confidence should come from what can be inspected, not from the fluency of the output.

Why it matters

AI can produce plausible but wrong answers. Trust needs evidence that can be checked.

Useful question

What would I need to see to trust this conclusion?

Practical example

A forecast predicts a 20% increase in leads. The reviewer asks to see the underlying assumptions and data ranges before accepting it.

Reflection — workflow review

1. What evidence do we currently ignore because the output looks good?
2. How might we make evidence visible by default?
3. What would change if every conclusion had to show its work?