Run reference

Check data rules, column definitions and run settings here. For a walkthrough, follow Create your first forecast.

Run settings

Use JSON configuration when needed

For advanced setup, you can enter settings as text in JavaScript Object Notation (JSON). Load a template with the fields Forecast accepts. Remove any fields marked as rejected. Fields marked as ignored won't be used, but they won't block submission. If the template can't load, use the starter JSON. Forecast checks it when you submit.

Choose the result you need

Choose Train and predict to train a model and make forecasts in one run. Choose Train only to train without a forecast. Use Prediction to forecast from a trained run. The form shows only the data sources, columns, dates and earlier runs needed for your run type.

Customize only what you need

Select the optional areas you need. If you hide an area, the form keeps its values but leaves them out when you submit.

Enter column lists

Enter one column name at a time. Press Enter or type a comma to add it. You can also paste a list with commas or line breaks between names. Remove any names you don't need.

Reuse configuration

Reuse configuration starts a new run with settings from an earlier run. It leaves out fields Forecast no longer supports and lists them for you to review. The saved run stays unchanged. Check access details, dates and output locations before you submit.

Create an experiment during submission

A new experiment is selected for you when you create it. When you reuse or build on a run, the form may suggest its experiment. You can choose another.

Schema terminology

Term	Requirement	Examples	Meaning
Target columns	Required	sales temperature, humidity	The columns with the values you want to forecast, such as sales. These are called targets.
Timestamp column	Required	pos_date utc_time	The column with the date or time for each row.
Identifier columns	Optional	product_id state, city	The columns that tell one series from another, such as a store and product ID.
Static columns	Optional	product_category, product_height, product_weight, product_depth country	Columns with values that stay the same for each series, such as a product category.
Known columns	Optional	discount_rate season	Columns with values that change over time and are known or planned for the forecast period.
Unknown columns	Optional	price, GDP wind_speed	Columns with values that change over time but aren't known for the forecast period.

Data requirements

Use long format: one row for each series at each date or time. Include its target and other input values in that row.
Target values must be numbers.
The timestamp column can hold a date or a date and time. Use the same format in every row. Prefer the ISO 8601 standard, such as 2026-01-31 for a date.
You can leave identifier columns blank if your data has only one series.
Static, known and unknown columns can hold categories or numbers.

Orchestration terminology

Term	Meaning	Details
Pipeline	A set of steps to train a model, forecast, explore past patterns or check results.	Training: Uses past target and feature data to create a trained model. Prediction: Uses a trained model and future input data to make forecasts. Complete: Trains a model and makes forecasts in one run. It uses your past and future data. In the guided form, choose Train and predict. Past explainability: Explore historical patterns. These results do not explain the forecast model's decisions. Future evaluation: Compares forecasts with actual outcomes. It can also show how input features affect the results.
Run	One attempt to run a Forecast pipeline.	A run can be started directly or created as part of a project.
Experiment	A collection of runs.	Groups runs that use related data or settings.