Prediction Game

Outguess

Try to be unpredictable. A tiny AI predicts your next key press by learning patterns in this session.

AI guess

Leaderboard and large tally show blind trials only. Models learn from the whole session; these are separate score slices, not independent experiments.

Blind: 0/0 correctHint visible: 0/0 correct
AI (Frequency) 0 / 0 (0%)
0 / 200 trials

Arena

Each tap is one trial. Predictors guess the next tap before it happens.

tap or press a key. try to be unpredictable.

Leaderboard

Rolling 50-trial accuracy. Bits = bits saved per trial vs random.

Randomwarmup

Coin flip baseline.

acc0%
bits0.00
Frequencywarmup

Bets on the most-common symbol so far.

acc0%
bits0.00
Markov-1warmup

Looks at your last symbol to guess the next.

acc0%
bits0.00
Markov-5warmup

Looks at your last 5 symbols.

acc0%
bits0.00
PPM-8warmup

Prediction by Partial Matching, max order 8 with escape blending.

acc0%
bits0.00

Entropy

Conditional bigram entropy of your input. The lower it goes, the more patterned you are.

bigram H(X|prev)0.00 bits
0fill 0%1.00 bits

Why?

Internal state of the leading predictor (-).

Play 20 trials in this scoring mode to inspect its leading predictor.

Why this works. Markov and PPM predictors can learn repeated key-tapping patterns. This demo does not claim a population-wide accuracy. Hide the AI guess to play a fair test of your own randomness, and watch the lifetime tally to see how the leading model is doing across the whole session.