There are not supposed to be good and bad relays on Nostr, at least not per se. If Nostr is working as envisioned then asking which relays are good should be like asking which numbers are good for an IP address. Relays are not supposed to be smart enough to really differentiate themselves in "cool" ways.

There are supposed to be just the relays you are using and the relays you are not using. And what differentiates them in general is limited to mundane things, who owns them, what region are they in, and so on.

Nostr only really scales if the load is spread across hundreds, if not thousands—if not tens of thousands–of relays. You can't spread the load that way if people clump, so motivation to clump needs to be removed. (Obviously if certain relays are seen as "good" then people will clump there, and at present over half of all users are clumped around 5 or so relays.)

There's argument to be made that (a) this vision is already damaged beyond repair, and (b) Nostr is showing more potential in terms of things less reliant on relay spread, such as interoperability—and as such Nostr should pivot.