Now the loop that does the actual work, and it has three steps that never change.

• Before: write what you expect to happen. Specific enough to be wrong.
• During: do the rung as defined. Without the crutch, without the phone, without leaving at the worst moment.
• After: write what actually happened, and answer one question — did the thing you predicted occur?

Why is the writing not optional? Because without a record, an afternoon that disconfirmed a prediction gets remembered as I got through it, but it was awful — and awful is compatible with the belief. The prediction was that you’d collapse. You didn’t. The record is what makes that visible, and memory will not do it for you.

Now the part that surprises people: you are not trying to make the fear go down.

This way of practising is built on something called inhibitory learning: what changes is not the strength of the fear during the exercise but the violation of what you expected.¹⁰ A practice where you were terrified throughout and the predicted disaster didn’t arrive is a good practice. A comfortable one where nothing was at stake teaches nothing at all.

This matters practically, because the old way of thinking — stay until the anxiety halves — turns every practice into a wait for a feeling, and makes a session where the fear stayed high look like failure. It wasn’t. Ask instead: did I find out something I didn’t know?

Three ways a practice quietly teaches nothing — all of them common, and none of them obvious at the time:

• You left at the peak. Understandable, and it’s the one arrangement that reliably teaches the wrong lesson — the fear was rising, you left, it stopped. That sequence looks exactly like leaving being what saved you. This is why Lesson 12 insists a rung ends where you said it would rather than when it gets bad.
• Nothing was at stake. A rung you were confident about produces a pleasant afternoon and no learning. If you’d have predicted it would go fine, it wasn’t a test.
• The prediction was unfalsifiable. It’ll be horrible can’t be wrong. I’ll have to leave within five minutes can. If you can’t imagine what would count as the prediction failing, rewrite it before you go.

Then vary it. Learning that stays attached to one setting is fragile, so spread the practices across contexts, duration, emotional intensity, and predictability.¹⁰ Not the same shop at the same time every Tuesday. Different shops, different hours, different days, sometimes when you’re already tired.

And when a rung is genuinely solid, put two things together. This is called deepened extinction: two feared elements at once, so that when nothing happens the mismatch is much larger.¹⁰ Here that means combining situations — the supermarket and alone and at the busy hour — rather than combining sensations, which is the part this program leaves out.

After a handful of practices, there’s something worth looking at: how many times you predicted something specific and it didn’t happen. Not a score, and not permission to move up — just the count.

The number that matters most in it is how many you did without a crutch. Those are the only ones with nothing else to credit.