Thank you for taking the time to explain your thoughts.
I first wrote all my queries in desktop, but when that started to grind to a halt, I decided I'd better research other ways. For sure, my queries are probably a bit rough and would make some people cry (!), but my dataset is tiny compared to some.
Thankfully, I already did what you suggested and copied/pasted queries from desktop via the advanced editor, and only used the dataflow interface for minor tweaks.
Previewing the query in service is nigh on impossible (it times out after 10 mins), and I put that down to sharing a server with many other tenants. When there are errors (I'm told there are errors, but it's impossible to see what they are!), I hook back in via desktop to see if the errors are obvious there, e.g. double data type (my lingo is probably incorrect). More often than not, it all looks ok to me.
What is particularly interesting in what you say is how the tardiness of validation seems to have fallen off a cliff in the last week or so; that gives me a ray of hope that it can/will be better than what I have now.
Thanks again, Robert.