how do you bypass it? didn't know its a thing

Hard reboots don't damage anything but the OS state, and even that has been protected against the last 2 decades.

The weak must perish.

This happens, too. A burst of local brownouts killed an OPNsense firewall of mine once. (The hardware was fine. The data got wrecked.) The firewall would boot, or almost finish booting, and then another brownout would hit. After the third or fourth cycle, the poor thing was toast.

I should thank the power company for free resilience testing.

Not sure about that. When I first installed Gentoo with OpenRC everything went fine, just the boot/shutdown times were a bit slower than with SystemD's when it didn't do this shenanigans. But when I started reading about the SystemD hype I decided to try it and it did this at random - and got fed up so went back to OpenRC and never experienced this again. Maybe a misconfiguration on my part or something but I didn't do something out of what the manual said about how to set it up.

If it was really the program's fault, why doesn't it happen when I switch to sysvinit? (or maybe openrc, I haven't tried that one)

That's not really systemd's fault though.

Oh, but it is. Upstart and sysV will both assassinate everything after a point. This bullshit is entirely new, either by Systemd sucking, or Systemd allowing this shit (thus, sucking), where neither thing happened beforehand.

I admit I rage when it's starting or stopping and can't win for all the losing -- especially because "startup time" was the the main (unfounded) feature on the banner they waved around.

I work with sysV and Systemd and shut each down on a regular basis, via tooling, for stuff like patching, so it's often a horse-race to offline. The difference is often negligible, for new machines or established boxes with apps, and the main difference is sometimes Systemd fails its YOU HAD ONE job from time to intermittent time.

I work in a team where Unix and Linux is ~15 very experienced admins, so don't fret over the skills and experience question or the scale question.

What I would call out systemd for is that it can make it difficult to know precisely what thing it isn't happy with. Abstracting a mishbehaving process to "user session" when it could call out the offending process by name.

Similarly, when it's trying to decide the network is 'online', and by all appearances the network interface was configured, but systemd isn't so sure, sometimes it's hard to tell why.

A lot of the bad experience with Windows similarly could be attributed to something workable, but the design obfuscates the underlying behavior. Systemd has been bringing some of that sentiment over, and that can be more aggravating.

I haven't gotten a startup wait in over 7 years but if I got it now I'd be pretty miffed.