I am also frustrated by the lack of a simple, local time synch protocol.  It could be used in many, many places.  For example, I had to replace the battery in my car recently.  A few minutes after starting it back up, the built-in GPS obtained lock, and knew the current time to tens of nanoseconds.  Six inches (15cm) away, the dashboard clock was still blinking "12:00".  Ridiculous!

I'm considering a 2-tiers approach with a local DS3231, a MPU with I²C and serial, and a GPS receiver that is turned on every week or so to catch the time...

But GPS does not work well indoors and not at all in my place so a loooong wire is necessary to reach the closest window.

Why not a RPi / GPS / WiFi combo server (a few watts) near a window, and an ESP8266 client that wakes periodically and uses SNTP to query the server. You could standardize on an "open" WiFi SSID and password just for this service - many access points allow multiple SSIDs (I have five in my house) .  You could make sure that traffic on that SSID is *only* ever routed to the time server, so nobody sneaks on your network (except to receive the time, if they want).

No, for multiple reasons.

NTP has been used as amplification relays for DDOS, it only moves the problem around and... not everybody has Internet access everywhere. So it's not the most cost-effective.

My favorite solution would be some DCF77 but it's not a worldwide standard and receivers are harder to find than 20 years ago :-(

If you run an NTP server on the RPi, you don't need Internet access at all - it gets the time from GPS and serves it to local clients over WiFi.  You're stratum-1. You do need local WiFi access, though, which may not be available.

If your network is routed/filtered so that time clients can only get to your local time server and vice-versa, there's little danger of DDOS attacks, especially if your time server isn't routed to the Internet at all (gets time from GPS).  But, you have to configure your network correctly :-)

I have two WWVB-synchronized watches that never set correctly.  I have always wanted to build a WWVB-emulator that produced a 60kHz time signal (it's well-documented) inside a metal enclosure I could store my watches in at night.  It would get accurate network or GPS time and make a "fake" local 60kHz signal (well-shielded to prevent interference to other devices). Add an efficient LED lamp inside to charge the batteries (the watches are both solar), and it works like a watch winder for automatic watch movements.

Don't know about the best method, but just sharing something semi-relevant I recently read.

A regular tuning fork have a tempco of about 86 ppm/C. That is incredibly low considering a ceramic resonator is about 3000-5000(!) ppm/C and it is just slightly worse than a crystal at 20-50ppm/C. This for just being a large fsking piece of metal...

I'm tempted to build a real 440Hz tuning fork oscillator just for funs when I get my lab back ^_^

Won't you quiickly get tired of the constant tone ? :-D
Furthermore, according to http://www.ieee-uffc.org/main/history-marrison.asp there are tuning forks in alloys with almost no tempco...

Already got tinnitus since a few years back so there's already a constant ringing in my ears but at a very noisy about 11KHz frequency.  But a tuning fork that is not put in contact to a resonating object is relatively quiet, so maybe it could be usable even in a home environment, especially if put under a glass dome.

Nice pdf.... I guess that having special alloys and multi-material tuning forks would be more than just *slightly* more expensive than the regular ones. I guess that the low frequency models also are quite expensive and much harder to find than the regular "musical tuning" models.

Still it would be cool to make a tube-based tuning fork oscillator and have mounted on a nice hardwood base and some brass knobs and stuff to make it "steampunkish" (even if tubes are a bit post-Victorian)....

They used tuning forks at one point:

https://monochrome-watches.com/watchtime-wednesday-history-tuning-fork-watches/