CIRCUIT is a predictive switchboard for your real area. Type your ZIP and it maps your actual neighborhoods, forecasts traffic, ETA and ride prices for each, and runs a reputation market where a machine-learning model and the crowd compete — then folds the crowd's wisdom back into the model. Live Waze traffic sits one tap away.
1
Your real map + the model. Your ZIP is geocoded into real neighborhoods on a live map (street / satellite / Waze). An online-trained model — it updates its own weights every round — forecasts each area's traffic, ETA and price. Watch its error fall on the learning curve.
2
The crowd forecasts too. You and 11 bot forecasters (each with a style — Momentum, Contrarian, Insider…) place predictions. A reputation-weighted average becomes the "crowd call" — sharper forecasters count more.
3
Reputation rewards accuracy. When the truth lands, everyone is scored. Beat your peers → reputation up, tier up, bigger payouts. Miss → you fade. No one can be perfect: real outcomes carry noise.
4
Feedback makes the model smarter. The reputation-weighted crowd call is injected as a second training signal. Toggle it below and watch the model's accuracy genuinely change — that's the core idea, live.
5
Incentives adapt. The system pays bigger bonuses for forecasts in zones where it's most uncertain (active learning) — it literally pays to be told what it doesn't know. Those zones get a ★.
6
The switchboard optimizes. Using the blended forecast, it routes capacity toward predicted hotspots before congestion hits — easing ETAs. Hit Inject Event (accident, storm…) and watch a neighborhood spike on the map, then recover.
Try the core experiment
Crowd feedback → modelWhen ON, the crowd's wisdom trains the model. Turn it OFF and the learning curve gets visibly worse.
Adaptive incentivesWhen ON, payouts steer toward high-uncertainty zones (active learning).
Honest note
The map, your location, and the live Waze traffic are real. The forecasts run on a simulation seeded by your real area (time of day, neighborhood type, demand) — they are not trained on a live traffic feed, so treat them as a smart model of the idea, not real-time routing. The machine learning itself is genuine: its error visibly drops and the crowd-feedback loop measurably changes it. Prices and ETAs are simulated — not real pricing, navigation, or financial advice.