Spigot 1.13 with Windows 10, no plugins or a plugin. Latest spigot at the time of writing.

There was one time when server reached 15 TPS based on Aikar's timing report, and it might happen again when server has many players online. I'm hoping to get accurate results thereafter.

Thank you for checking it out and fixing it. I appreciate it

TPS tracking has been fixed and tested.

Given your server is running at close to 20tps you may not see a difference, but maybe you will (at least the * will probably be gone).

Actually the TPS tracker is wrong, fancy that.
I don't think it will change your case however

Your timings report shows 19.81 tps average - i.e. no lag. Your server load is less than 75% indicating that also any lag spikes will be caught up and your TPS will not drop appreciably below 20.

If neither average nor ewma show your “lag”, what measurement do you expect will? Should we be showing standard deviation of tick time or something?

If your server lags for two ticks (say 300ms) and then runs the next 18 ticks within 700ms, tps will still be 20, regardless of whether it is an ewma, sma, or average.

The tps command is meant to give a quick overview of your servers general performance.
It can’t possibly account for lag like that mentioned above, that is something that has to be examined via a profiler.

I will double verify that /tps is reflecting accurate information, but otherwise this isn’t a bug and I think you’re just misunderstanding what TPS is capable of representing.

I'll report back in an hour or so after restarting my production server with latest spigot. The production server runs on Ubuntu 18.04 on OVH and had the same (yet more extreme) issue.

Edit:
Timings report of production server: https://timings.spigotmc.org/?url=lodidoyotu
But the /tps command still responds with *20 for all three.

I have another server on the machine, and that one gives values between 19 and 20. This secondary server has less plugins (and does not run CoreProtect).

Therefore and interestingly, this issue occurs when there's more plugins (or specific plugin with some event?) on the server.

This seems to be unrelated to the OS as I got to test it on both Windows 10 and Ubuntu 18.04.

The EWMA of TPS would be acceptable if the value reflects the real state of the server's TPS, but having it at *20 when players are experiencing eating lag (eating for longer than normal) or block-breaking lag (block reappearing before breaking again) is pretty questionable and unreliable.

TPS command uses an exponentially weighted moving average
Timings uses a simple average
Essentials /gc command uses a simple moving average.

Obviously all of these will produce different answers.
I will double check that /TPS is reporting values correctly still (it certainly reported less on spigot craft during testing).

But I doubt this is a bug.

Reporting the "true" TPS would be useless as it would only show lag within the last second, which is useless.

EWMA as currently used is imo the best because it gives a general overview whilst prioritising the last minute or so.

Your timings paste also shows tps as effectively 20.

19.94 is within the margin of error on your OS scheduler.