LLM-readable documentation index
Lab

Datasets

Collect labeled outputs for function comparisons

Hand off to an LLM

Use Datasets to group outputs you have confirmed or corrected into a reference Golden Dataset. These can later be used in Comparisons to evaluate accuracy across different functions or different function versions to see how different changes affect the results.

Create a dataset

First, ensure you've labeled a sizable set of outputs in the Corrections page. Then select New dataset, give it a name, and choose the outputs to include.

Under Functions, you can include labeled outputs from all versions of a function or from its specific version. Use Individual events to include specific outputs by reference ID. You can combine both sources in one dataset.

New dataset panel with function versions and labeled event sources

Inspect a dataset

Open a dataset to see all its included outputs. Expand a row to see the expected content and source function version of the given output. Ensure that the labels are correct and the examples cover the inputs you want to evaluate.

Dataset details with an expanded row showing corrected content

On this page