Learn Better With AI

Methods · Generative practice & retrieval

B28

Criterion-based exit

Keep practising until you hit a self-set accuracy and confidence bar, not until the worksheet ends or the clock feels done.

When to use this
When checking yourself
Learning impact
Strong change
Learning evidence
Credible with caveats
Where you are with the topic
Checking readiness
How often to use it
Build and keep as a habit
What AI is doingRemember your plan across days

When to use it

  • You stop because you finished ten questions, not because you can do them.
  • Time-boxed sessions end mid-mastery and you call it complete.
  • You want a clear exit rule: for example, 4/5 correct with confidence ≥4.
  • Easy early items trick you into quitting before the hard ones.

What to ask AI

Type something like: “Exit criterion: 4 out of 5 correct with confidence ≥4 on this skill. Keep giving unmarked items until I hit it or I call stop. No answers until I rate confidence.”

Set both accuracy and confidence if overconfidence is your pattern.

Ask it not to congratulate you for finishing a fixed count alone.

How it helps

Fixed item counts measure time spent, not readiness.

A criterion ties exit to performance you can defend.

Adding confidence stops you from clearing a bar with lucky guesses you do not trust.

The AI keeps serving items; you decide honesty of ratings and when to stop for fatigue.

How to run it

  1. Write an explicit exit criterion before you start.
  2. Attempt items; log correctness and confidence.
  3. Continue until the criterion is met or you hit a planned fatigue stop.
  4. If you stop early for fatigue, note that the criterion was not met, do not pretend it was.
  5. Next session, resume until the bar is actually cleared.

Example exchange

You

I want mixed chain-rule practice with a clear exit: stop when I get 4 out of 5 correct with confidence at least 4. Keep going until I hit that, one item at a time.

AI

Item 1: differentiate y = (2x+5)⁴. Answer and give confidence from 1 to 5 before I mark.

You

8(2x+5)³, confidence 5.

AI

Correct. That is 1 correct so far toward your 4-of-5 exit. Item 2: differentiate y = sin(4x). Answer plus confidence before I mark.

Copyable prompt

Skill: [ ].
Exit criterion: [e.g. 4/5 correct with confidence ≥4].
Keep giving me unmarked problems until I hit the criterion (or I declare fatigue
stop). Require confidence before marking each item. Do not treat a fixed count
as success.

The Tell

You will know the method has flipped when you quietly lower the bar mid-session (“3/5 is fine”) after a miss, or stop at a round number of items with no criterion check.

Then exit was mood, not mastery.

Write the criterion first, and if you stop early, label it fatigue, not done.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

Mastery-oriented practice uses performance criteria rather than seat time. Pairing accuracy with confidence connects performance to calibration.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

AI can enforce item flow; it cannot force your honesty. Criterion-based exit fails if you fudge ratings or redefine success after errors.

Related methods