From description and code, it looks like the local file is special in that it is considered global, even if it is invalid - if that's wrong, the following is moot:
It seems inconsistent and afaik also will result in an inflated global count: We don't include invalid files in the local count (at least used to in v1). What's the reason to mark an invalid local file as global, instead of not having a global one like in the case where there's no local?

What's the reason to mark an invalid local file as global, instead of not having a global one like in the case where there's no local?

Good question, and that was also my gut feeling & first fix attempt. We do use the local-is-global-is-invalid scenario in one case, namely receive only folders with local changes. Their behaviour seems to depend on this enough that a couple of tests fail if we don't do it. I didn't want to get too deep into those weeds right now.

It does in any case end up consistent (I think; from what I can see), as we do exclude files with the ignored flag when doing the local counts etc. At least when I test...

Ok, I understand now how this ends up with consistent local and global counts: We do consider local flags in counts, but we don't consider the invalid bool. That's why marking an invalid local file as global is fine, it has local flags, while adding a invalid remote file inflates counts.
Feels a bit icky to go to this half-way state, where some invalid files aren't global, and others/local ones are. We could add a column for invalid to counts to fix the regression/get back the old behaviour until we can mark local invalid as not global too.

And don't we need a migration to remove the global flags with this change from the affected files?

Also long-term/conceptually, I think it's nice-ish to keep marking invalid files as global. That e.g. allows to find all files that are "globally invalid" easily, so we could also drop them from the DB. Given we now clean up old deleted files (still a bit worried about this one :) ), also cleaning up old invalid files seems quite sensible.
And the whole invalid/local-flags split feels a bit inconsistent resp. error prone in handling, I am pondering getting rid of the invalid flag internally and adding FlagLocalRemoteInvalid (converting from that to the invalid flag in the from/to wire functions). Then counting would also work fine, as we already count by local flag.

And don't we need a migration to remove the global flags with this change from the affected files?

... in principle, yes. But we're talking about a few hundred RC users so far. I was going to say reset deltas if this is an actual problem.

I am pondering getting rid of the invalid flag internally and adding FlagLocalRemoteInvalid (converting from that to the invalid flag in the from/to wire functions). Then counting would also work fine, as we already count by local flag.

This sounds like it has merit. Let me look into that (or you can, if you get to it first ;)

This sounds like it has merit. Let me look into that (or you can, if you get to it first ;)

I am already 90% there, started before I saw your message - hope we didn't do it twice now. Still looks like it has merits to me, just a bit a pervasive change :)

Here it is. However take with a good bit of salt, I just pushed this to pass tests without stopping to think overly much along the way, good chance there are some questionable (at best) decisions in there:
#10170