Modification is defined exactly the same in the AGPL as in the GPL. I don't think putting a reverse proxy in front of an AGPL app means much of anything in terms of either the letter or the spirit of the license.
Mongo's separation of Apache-licensed drivers from the server is about making it so that applications don't have to link to the AGPL'd server code, thereby avoiding "modifying" it. Without a modification, the network interaction section isn't triggered at all.
Still, my main point was that the route taken by Mongo is intended to remove this kind of uncertainty: normal applications using it don't modify or include any AGPL code, so the question of just how far "remote interaction" goes never comes up.
They must've heard about the problem. Why won't they make a statement about intended reading of AGPLv3 and publish AGPLv4 that has improved wording that avoids any possible ambiguities?
why FSF isn't stepping up and resolving this issue once and for good?
Because they don't have standing. They're not a judiciary, so they can't establish precedent - it take a court to do that.Sure, they could change the wording, but unless there's a court verdict to establish precedent, there will always be ambiguity.
To put it in other words, clarifying won't hurt, even if it could have no real effect.
But, yeah, the real way to fix is AGPLv4. And FSF surely can do this.
In the past MongoDB has argued that using their AGPL database without providing source code is fine, Neo4j has argued the opposite. As strictly interpreted it seems like as long as you don't expose the AGPL app to end users you're not in violation. Maybe the license authors intended this, maybe they intended "if you use an AGPL application that in any way affects an end user's interaction, you must provide the source to that end user."