Behavioral Experiment Platform

100numbers

A simple experiment in human randomness. Enter digits, lock a blind prediction, and compare the model against your next taps.

Three steps. One hundred numbers.

The flow stays simple: capture a run, lock a prediction before the reveal, then inspect how well the model anticipated your next digits.

01

Enter a run

Type digits in any order at your own pace. The goal is to capture your natural sense of randomness.

02

Commit the prediction

Once you reach the run target, 100numbers locks a prediction file for the next 10, 100, or 1000 taps.

03

Evaluate fairly

After you finish the evaluation window, compare the predicted sequence with your actual digits and resulting accuracy.

Human randomness carries structure.

People tend to avoid repetition, over-correct, and reveal small habits when trying to look random. This project turns those habits into measurable sequences.

Each run is treated as a separate experiment. You provide the digits, the platform locks a blind prediction, and the result becomes a transparent record you can inspect afterward.

Run preview
Your input
3
7
1
4
9
2
8
0
5
6
Prediction
3
7
1
5
9
2
9
0
5
6
Match
8/10 positions matched in this sample window.

Your numbers already leave a fingerprint.

Start a run, let the system commit a prediction, and see whether your “random” choices stay random under measurement.

Blind predictions for human generated sequences.

Sign in with email and password or use Google. Your session token stays in this browser so you can continue your runs and inspect previous evaluations.

What you can do here
Collect new digit runs in real time.
Commit prediction files without seeing the output first.
Review accuracy, prediction comparisons, and model metrics after evaluation.

Sign in

Sign in with email/password or Google. Your session token is saved in your browser.

or continue with Google

Minimal context for your future runs.

Everything here is optional. It just gives you a lightweight profile summary inside the app and stays attached to your account in this browser.

Stored fields
Country or region.
Primary language.
Place you grew up and a lucky digit, if you want to add them.

Quick profile

Optional. Keep it minimal or skip it entirely.

Mode: collecting
Keyboard input works too.
Prediction commit
Lock a blind prediction for the next 10, 100, or 1000 taps once your run has enough data.
Locked until target is reached.
Personal history digits Global history digits Checking model availability…
User
guest
Total digits
0
All time
Saved profile fields
Previous predictions
Model observatory
Committed telemetry

See how the model performs.

A clear view of every locked prediction—how often the model is right, which version performs best, and whether results hold over time.

Weighted accuracy Primary
Random baseline is 10%
Evaluated commits
— of all commits evaluated
Model outputs scored
Selected + background positions
Total commits
Locked prediction windows
Performance over time

Accuracy trend

Waiting for data
No data yet.
Model comparison

Versions in play

Fair head-to-head test

Same-run model comparisons

One future digit window per row

Your selected model remains the official prediction; the other models are measured privately against the exact same digits.

Collection context

How the digits were entered

Newly collected data

Input and device labels are recorded from this release onward. Earlier digits remain unclassified.

Input method

Device type

Audit trail

Recent evaluations

Newest first
Email

hello@100numbers.site

Horizon
Accuracy
Run summary
Most tapped digit
Model used
Prediction comparison

Green rows matched. Red rows did not.

Position Predicted Actual Result