A forecast is a set of estimates made from a particular history, model configuration, and cutoff. The median is useful, but the surrounding range helps you ask better questions about uncertainty.
Start with the median, then widen your view
P50 is the model's median estimate at each forecast timestamp. P25 and P75 describe the middle of its predicted distribution; P01 and P99 describe more extreme estimates. These are model quantiles. Their labels do not prove that future prices will fall inside the range at the stated rate.
Compare the same timestamp
Match the observed price to the forecast date and horizon you are reviewing. A terminal P50 seven days ahead answers a different question from tomorrow's P50. A range that widens over time indicates increasing model uncertainty; it does not establish that a trade has an edge.
Check which models support the range
Inspect the models and effective weights. A model may support central quantiles while another supplies the extreme tails. Quantura's game ensembles exclude models without native P01/P99 support from those tail calculations.
Turn the output into a research question
Write down the outcome you want to test, the forecast cutoff, and the evidence that would change your view. Save the result before reviewing later observations. Compare several forecasts with realized outcomes; avoid drawing a conclusion from one successful example.
Updated September 26, 2026.