We have found an apparent issue/bug in dataflow refreshes, where linked dataflows have different 'enhanced compute engine' settings.
To deal with a previous dataflows bug (now resolved), we disabled the enhanced compute engine for some of our dataflows. When the bug was resolved, we set the enhanced compute engine setting back to 'optimized'. We actually did this for some dataflows and not others, in a set of linked dataflows in the same workspace which use linked and computed entities.
We have now found that the data/entities in the dataflows where enhanced compute was still disabled have not actually refreshed at all - even though the 'Refresh history' shows a long series of 'Completed' data refreshes. The data also seems not to have updated even after a 'On demand' refresh. Essentially the data/entities have been 'stuck' in the same state.
We have now resolved this by setting the enhanced compute engine to 'optimized' on all of the dataflows.
I have found nothing in the documentation for dataflows or enhanced compute engine which explains that the enhanced compute setting needs to be the same for all related/linked dataflows, nor can I see any good reason why this would be necessary. So this appears to be an issue/bug which Microsoft need to resolve.
Per the latest reply from the Product group in the ICM for the similar, they are working on fixing the bug. But it still need about one month. Please be patient for the fix. Thanks for your understanding!