A spreadsheet should help you follow the research, not hide how it was produced. A clean time series and an explicit export make it easier to compare model estimates with later observations.
Make the input unambiguous
Use one timestamp column and one numeric target column. Sort timestamps, check duplicates, state the timezone, and verify the frequency. Price quotes and contract probabilities use different units. Keep the source and any adjustment choices with the file.
Choose a question and horizon
In Forecast, select your CSV columns, models, quantiles, and horizon. Inspect the observation count before running. A longer forecast or another model cannot repair missing or incorrectly labeled inputs.
Keep the output together
Save your forecast to Requests, then export the quantile table for a spreadsheet or another analysis tool. Preserve the source, input cutoff, model choices, and forecast timestamps alongside the predictions. Developers can use protected APIs when the same workflow needs to run in their own tools.
Review without changing the original question
Add realized observations in a separate column. Compare them at matching timestamps and record both useful forecasts and failures. Keep the original result so a later model refresh does not replace the evidence you intended to evaluate.
Updated September 26, 2026.