The Lucid Calibration Test
How right are you, really?
3 minutes. 12 questions. One uncomfortable number.
Free. No signup. Mildly humbling.
This is Lu. The glow is your stated confidence. Reality decides what happens to it. The test measures the difference.
Sol, the coach
Every other AI tells you what you want to hear.
Chatbots are built to validate. Agreement keeps you talking, so agreement is what they sell. Lucid’s coach is built on the opposite principle: he will never tell you that you’re right. He asks the question you skipped, then reality does the grading.
Not validation. Measurement.
How the training works
Metacognition, the skill of thinking about your own thinking, is built the way any skill is built: repetition against reality.
Predict what matters
Real predictions about your actual life: the deal, the deadline, the decision. Concrete, measurable, and yours. Not trivia, not hypotheticals.
Reality gives the feedback
When the moment arrives, you record what actually happened. The feedback comes from your own experience, not from an algorithm's opinion of you.
Measure the gap
The distance between how sure you were and how right you were, on the challenges that are actually yours. That gap has a name: calibration. It is trainable.
What the test actually measures
Twelve claims about the decisions people actually face: money, work, relationships, risk. Each takes two answers: your call, and how sure you are, from 50 (a coin flip) to 99 (certain). You are never graded on being right. You are graded on whether your sureness matches your hit rate. Wrong at 55 means you didn’t know, and you knew you didn’t: an accurate read of yourself. Wrong at 95 means you didn’t know that you didn’t know: the blind spot. That match is metacognition, and it is trainable.
This isn’t a quiz. It’s training.
Calibration, knowing how much to trust your own judgment, is one of the few thinking skills with decades of evidence behind it.
0 yrs
of published calibration research behind the method
~0
scored judgments turned overconfident people measurably calibrated
<1 hr
of training sharpened real-world forecasts for a full year, in a randomized trial
Lichtenstein & Fischhoff (1980) · Good Judgment Project RCT (Chang et al., 2016)
The record
In fifty years of studies, people who say they’re 90% sure are right about 60–70% of the time.
The test measures where you stand. Most people have never seen their number.
Built to make itself unnecessary.
Most tools measure success by how much you use them. Lucid measures whether your judgment holds up when the prompts are removed. That principle is written into the system: when engagement and learning disagree, learning wins.
Training, not entertainment
Nothing here exists to keep you scrolling. Every mechanic serves a repetition: a prediction made, resolved, or learned from.
Misses are data
Every wrong call lands as a question worth answering, never a loss to be ashamed of. The miss is where the learning is.
Graduation is the goal
The finish line is needing the tool less: judgment that holds up on its own. We measure for that outcome and build toward it.
Then the training teaches you to move it.