A fixture for your application. Start with 1,000 e-commerce orders, weighted statuses and missing ratings. Review 25 rows, then adapt the schema to your tests.
A controlled evaluation scenario. Inspect 10,000 transactions with channel-dependent rare labels. Use the simulation to check splitting, ingestion and metrics; it is not evidence of real fraud-model accuracy.
A path for training-data extensions. Upload your own table or select the 3,000-row fictional loan sample. Download the same source CSV before starting analysis.

October 21, 2026Ground team

ImprovedDesign

A new mark, a warmer palette, and a little motion that makes Ground feel more like Ground. The new identity now runs throughout the platform.

A new shape for Ground.

One identity throughout the product. Shared marks and wordmarks replace the placeholder artwork across the app and public pages.
Motion with a consistent rhythm. The Contour pulse connects logo hovers and loading states, including dataset design progress.
Artwork that stays sharp. SVG logo and icon variants support light, dark and monochrome uses, alongside browser and app icons.

October 20, 2026Ground team

ImprovedProduct

Follow analysis and export progress across Ground, return to a clear results summary, and compare or restore earlier design revisions.

Progress that follows you. Leave the dataset page and keep track of work through app-wide job notifications.
A clearer next step. Open an analyzed dataset to its results and inspect suggestions about thin groups, rare labels or source-data issues.
Room to explore. Compare saved design revisions side by side and restore an earlier version when you need it.

October 20, 2026Ground team

NewDevelopers

Download the data, inspect the exact rules, and verify the output yourself. Worked examples now include specifications, seeds and file checksums.

ecommerce-test-dataseed 991

data.csv

data.jsonl

spec.json

manifest.json

The output. The rules. The seed.

Measured output. Inspect observed status frequencies, missing values and rare-label counts rather than relying on probability settings alone.
An inspectable recipe. Save the specification and seed with your fixture. Recreate a dataset through the API using the documented request.
Clear limits. Each example states what it models and what it leaves out, so a designed simulation is not mistaken for validated real-world behavior.

October 20, 2026Ground team

NewProduct

No file handy? A built-in sample lets you explore source analysis and the extension workflow with 3,000 fictional loan applications.

A useful starting point. Analyze 3,000 rows across 16 columns without preparing or uploading a file.
Inspect before extending. Review the profile and quality report, then choose how many rows to generate.
Free to explore. Sample analysis is free, up to three tries per account. Normal export pricing applies when you generate a file.

Worth a closer look.

Every update, in the open.