Most exploratory analysis does not need a published artifact: you want to check a hypothesis, debug a customer issue, validate a join, see what the data actually looks like. Playgrounds are the place to do that. The query runs against the same data sources your pipes do, with the same engine, but nothing is persisted unless you choose to promote it.

When a playground query proves useful, it can be turned into a pipe with a click, which is the natural path from exploration to production. Until that happens, the playground stays out of the workspace graph and out of version control.

Playgrounds are not a separate database. They are a scratch surface on the same workspace, which means access control, tokens, and data isolation all apply.