Depending on how you count "the NTP Pool" is either just me or it's all of us, hint hint.
I actually have made that (running FreeBSD/NanoBSD), but it still costs $300-$400 or so -- seems like too much for a hobby when just running ntpd on some linux box you already have is almost as good. Maybe for people who have a static IP but no server running 24/7? Seems like a small group...
A small computer with the appropriate serial port (Soekris box, for example - $200 - $250 with power supply and small compact flash) plus Garmin 18lvc with the appropriate cabling (~$100). Then you still also need an (extra?) static IP address and space near a window (as you said). Doesn't seem like a big market!
Maybe I'll do that in the future if we setup a separate DNS name for SNTP clients. For "ntpd" it's not currently practical. The ntpd developers are working on some new features that might make occasionally changing IPs work better with it, but it'll be a long time before they're widely deployed.
Being able to leave, as you did, is part of the point of the pool system. With the static lists of DNS names and IPs, there wasn't a good way to stop providing service again.
It is frustrating with the abusers, but getting that fixed is a parallel problem to providing service. With the pool at least we can spread the abuse out over thousands of servers rather than having a handful of hardcoded servers getting all of it.
As others pointed out, NIST operates a set of high quality clocks available via NTP, too. Last I talked to someone there and tried to estimate the pool.ntp.org usage the two systems got a comparable number of requests.
There are different tradeoffs to using NISTs servers and using pool.ntp.org.
A long cable.
An (expensive) way is to use a CDMA receiver. The CDMA protocol needs accurate timing, so it's included in those signals. It's not as accurate as GPS, but it can work in places with no "sky view" access.
Yes, if a USB receiver makes it accurate enough for the monitoring system then it's fine. (Though the monitoring system has been tuned to be stricter and stricter over the years).
However: if the USB receiver has more "jitter" than the other internet servers you'd be syncing from as backup, then there's not much point in having it.
We have a equal opportunity Calculus class -- it's fully integrated.