What does it mean that predictions test schemas?

If your schema is correct it should make accurate predictions about what will happen next.

Example: You believe that your colleague resists new ideas because she is risk-averse. If this schema is correct, it should predict specific behaviors: she will push back on any proposal that lacks precedent, she will favor incremental changes over large ones, and she will ask for data before committing to anything unfamiliar. You write these predictions down. Over the next month, you observe that she resists proposals from certain people but enthusiastically backs equally risky ideas from others. Your predictions failed — not because she is unpredictable, but because your schema was wrong. She is not risk-averse. She has a trust filter. The prediction failures did not just reveal inaccuracy. They pointed directly at the flaw.

Try this: Select one schema you currently hold about a person, a system, or a recurring situation. Write down three specific, observable predictions that this schema implies. Be concrete: what will happen, when, under what conditions. Then observe. Over the next week, track which predictions are confirmed, which are disconfirmed, and which you cannot evaluate. At the end of the week, score your schema: did it predict well, poorly, or ambiguously? If it predicted poorly, what does the pattern of failures tell you about where your model is off?