Quarter-Hour BTC Prediction Lab

Symbol Bootstrap history Prediction target Temperament: Balanced Froggy ← → Cautious

The engine evaluates the active 15-minute interval repeatedly. It freezes the first call that clears its time-adjusted confidence threshold. New calls are prohibited during the final 30 seconds.

Live price

—

Disconnected

Next boundary

—

Current interval decision

—

No trained model

Walk-forward accuracy

—

No scored predictions

Minute price and quarter-hour boundaries

Model state

Minute bars	0
Training examples	0
Scored forecasts	0
Coverage	—
Mean decision time	—
Brier score	—
Log loss	—
Baseline accuracy	—

Recent boundary predictions

Interval UTC	Decision time	P(above)	Threshold	Call	Actual	Correct	Reference	Outcome

Feature diagnostics

Feature	Current z	Weight

Calibration by probability bucket

Forecast bucket	N	Mean forecast	Observed above

Method and limitations

The model trains on simulated checkpoints inside historical 15-minute intervals. Features combine pre-interval context with partial current-interval price, volume, volatility, range, elapsed time, and distance from the settlement reference. The first live call clearing the adaptive threshold is frozen until settlement. Reported performance uses chronological walk-forward simulation.

Seven days provides only about 672 quarter-hour examples. That is enough to test plumbing, not enough to establish a durable edge. Exchange-specific price, latency, fees, market spread, prediction-market settlement rules, and regime shifts can eliminate an apparent advantage.