Try the following scenario from Falcon with an SQL connector like Postgres:

1. Create a query select 1 as a and save it. The result is as expected.
2. Update it to select 2 as a and save it. The result is as expected.
3. Update it to select 3 as b and save it. Note that the grid on the Plotly side still has the column named a and not b although the cell value in the row is indeed 3 and the metadata did get updated.
4. Update it to select 3 as a, 4 as c and note the pattern: still 1 column only.
5. Update it to select 3 and note the JSON error without indication of which processing-stage it came from.

A different set of examples is possible when the original dataset has more than 1 column and we try to update it with a dataset with fewer columns: more errors.

These errors don't appear in the webapp-based UI, presumably because it has a different set of API calls when registering a query change, which re-registers the columns?

Heads-up @scjody, this is likely a new blocker for 2.5.0.