Yeah, I was actually picturing such a feature with an existing database being implemented under the hood using a transaction block around the query input. This idea could also be extended to the most basic case of a completely empty database without modification, instead of just resetting the database, it would just use the same transaction logic.

The transaction would be left open until the user either decides to re-run the (presumably modified) query input (causing a rollback first) or to (presumably) end the session by downloading the database (causing a commit first). This behavior would only apply in the auto-resetting “fiddle mode”, of course.

Anyways, it was just an idea I was thinking about. I’m sure this idea is not flawless, and you probably have other, higher priority features in mind.