Then why does my email program reliably distinguish spam from ham without any server-side filtering involved?

If you have a client app that you think is reliably filtering spam, then to be blunt: you aren't receiving any spam, at least to first approximation. My stack of stuff on my three decade old personal account has a 95%+ hit rate (something I might naively be tempted to label as "reliably filtering spam"), and I see see more spam than signal in the inbox.

GMail, on the other hand, is damn close to 100%. And it does it through excruciating application of heuristics like "don't trust agents that don't set SHOULD headers".

I'm just speculating, but probably because you're on an email provider that isn't a big enough target to worry about the persistent threat-actor model.

Google is a big enough target to justify spending the resources on dedicated attacks against their infra. Other providers may simply get less spam because their email domain shows up less often in the sources attackers use to pick targets.