Every attempt in Brainback asks the student to rate their confidence 1–5 before submitting. This is not decoration. It is data for the single most useful meta-skill in learning: calibration.
Calibration is your ability to distinguish the questions you actually know from the questions you almost-know. Well-calibrated students study efficiently, allocate exam time correctly, and stop being ambushed by 'I thought I knew this' failures.
The Brain score uses a Brier-style calculation: we compare your confidence to your actual outcome across recall reviews. Perfect calibration means high-confidence answers are almost always right and low-confidence answers are almost always wrong. Miscalibration in either direction — over-confidence or under-confidence — is training data for future prompts.
Over-confidence is a study problem: you don't practice what you falsely believe you already know. Under-confidence is a test problem: you second-guess correct answers and change them to wrong ones. Both cost points. Both are fixable — but only if measured.
Confidence tagging is the smallest possible intervention that lets us measure this. It takes one click. It changes the study game.
Written by Brainback Research. Published Feb 27, 2026. Filed under Method.