Walk-Forward Results

Walk-forward analysis tests a model on periods that were not used to define it. Each row uses an earlier period for training and a later period for evaluation. Repeated success across several unseen periods is stronger evidence than a good fit to familiar historical data.

:

Model Window Lab

This secondary lab is restricted to models frozen into . It explains their historical unseen-window behavior and current scenario calculations. The authoritative round forecast-versus-actual comparison is above; runtime fallbacks and passive market counterfactuals shown below are not substitutes for a missing round ledger.

Forward estimate

Projected vs realized value by selected model

Dashed lines are the original forecasts. Solid points are recorded paper results where available. A labeled passive-price comparison is shown when a model does not yet have its own paper record.

Asset outcome lens

model outcomes vs asset buy-and-hold

Compare the selected models with the chosen asset before interpreting their portfolio projections. The chart places each Day 30 model estimate beside its recorded result, when available, and the asset's Coinbase buy-and-hold return through the current day.

Selected asset actual

market actual since baseline

This line shows what happened to the selected asset's Coinbase price after the starting date. It is the passive buy-and-hold comparison, not a model result.

Selected-amount calculation

Projected and recorded outcomes at

Every row below uses the same starting amount. Projected ending value is the starting amount plus the model's calculated change; it is not a balance observed growing from zero. The modeled cost drag is an assumption in the forecast calculation, not evidence that a larger account would trade more efficiently.