Note: I KNOW that documentation is seriously hard. But, one of the rebuttals of the "Systemd is complex and monolithic" assertion is "Systemd is fully documented, so you can replace any part of it with one that you've authored.". You can't honestly make that rebuttal if the state of your documentation doesn't support your claim.
Oh, you summer child.
In any complex software project, there's always a difference between the contract described by the system's specification and the contract described by the software as implemented.
The big difference between the explicit contract of the documentation and implicit contract of the code is that -in areas on which the documentation is silent- the contract described by the code is subject to change at any time, without warning. Raymond Chen [0] has spilled much ink over customers who have gotten badly burned by changes in the implicit contract some time after they made use of something that the code permitted, but the documentation did not describe as possible and/or supported.
[0] Microsoft most notable backwards-compatibility guru
That's exactly the reason why source itself is better than implementation documentation - because it is always "current".
If user behaviour doesn't change - why changes in implementation are wrong? Reimplementations can use the source code at one place in time and use that as a reference.
I think systemd has the misfortune/honor of touching decades-old subsystem interfaces. Any time anyone ever makes a change to something that's worked the same forever faces criticism. systemd's feature set is so attractive that it's getting/gotten adopted by the vast majority share of linux distros. The folks who didn't prefer it but thought that they could ignore it may have been surprised and vocal when it showed up in their favorite distro.
>undocumented 200+ line init scripts ... undocumented shell scripts
Maybe we've just traded those for undocumented C init programs?