A useful forecast begins with a dataset you can explain. Before interpreting the chart, check what was observed, when it became available, and how it reached the model.
Verify the source and instrument
An exchange stock, a CFD, and a prediction-market contract are different instruments. Check the provider and symbol. For Dukascopy quotes, retain the selected bid or ask side and the published price scale. Changing a provider can change the meaning of a series.
Read the cutoff
The last completed observation tells you what information was available to the forecast. A candle's start time differs from the time its closing price becomes available. Historical strategy research must also account for the time needed to produce the forecast.
Keep missing observations visible
Check observation counts and gaps. Weekends, market closures, and periods without genuine observations are not interchangeable with feed failures. Filling a gap introduces an assumption; silently filling it makes the output harder to interpret.
Separate a forecast from evidence of accuracy
Record the models, quantiles, and horizon. Then compare saved estimates with later observations or a chronological holdout. Repeatedly choosing a rule after seeing the same outcomes can make a result look stronger than it is.
For a game, verify kickoff with the provider. A market's listing date or settlement deadline should not be treated as the start of play.
Updated September 26, 2026.