This kind of difference can be surprisingly easy to miss because the same notebook may behave differently depending on how it is launched. In Databricks, compute type, runtime, permissions, parameters, and whether the execution is interactive or a job can all affect the result. Serverless compute can also have different API and library limitations, so comparing the actual execution environment is usually the first thing I'd check.

For the same reason, when experimenting with scripts in other platforms such as Roblox, I find it useful to check the exact version and environment before assuming the script itself is the problem. A resource like this Delta information page can be useful when comparing version or compatibility details.