> ## Documentation Index > Fetch the complete documentation index at: https://docs.promptlayer.com/llms.txt > Use this file to discover all available pages before exploring further. # Overview > Use Datasets as the versioned system of record for evaluation inputs, backtests, and batch workflows. Legacy Evaluations, Reports, and Datasets are deprecated for new workflows. Use [Tables](/features/tables/overview) for new evaluation, dataset, report, backtesting, and batch workflows. See [Migrate from Evaluations and Datasets](/features/tables/migrate-from-evaluations-and-datasets). Datasets are PromptLayer's versioned system of record for evaluation inputs and historical examples. Use them when you want a reusable set of test cases for evaluations, backtests, regression checks, or batch workflows. A **Dataset Group** is the container for one dataset over time. Inside that group, you edit a **draft** dataset, save numbered versions, and reuse those versions across evaluations. You can create data by uploading a CSV or JSON file, building from request history, adding individual rows from observability traces, or turning evaluation outputs back into a dataset for the next iteration. ## What you can do * Create datasets from files, request history, or evaluation results. * Add individual rows programmatically from observability traces via the API. * Edit draft datasets before publishing a version, including adding rows, renaming or deleting columns, updating values, and converting columns to JSON. * Organize dataset groups with folders, tags, changelog history, and archive controls. * Reuse a specific dataset version in an evaluation blueprint, or update a blueprint to a different dataset later. * Export datasets to CSV and use report outputs to seed the next round of testing. ## How it fits together 1. Create a dataset group and populate a draft. 2. Edit the draft and save a version. 3. Attach that version to an evaluation blueprint. 4. Run full batches, review results and history, and feed outputs into the next dataset iteration. ## Next steps Learn how datasets power evaluation blueprints, scoring, and batch runs. Build datasets and evaluations from code or CI workflows. Upload a CSV or JSON file to create a new dataset version. Build a dataset version from filtered request history. Add an observability trace or span subtree as a dataset row via the API.