There actually are pretty good measures of decentralization. You can look at it as graph and see what failure means (is it scale free?), and many others. Choosing the right metric is difficult though. People usually pick the one that says what they want to say. It's been the case with lightning for example.

Another problem with measures of decentralization is taking into account the dynamics. That can also be done (in this case fragility analysis). Yes, most posts are on one relay. The question is what would people do if it fails for some reason. Would they point to Peter's paper and say "haha, he told you so" or would they move on and use things differently, use other relays, mirror content, etc. Because that's what matters. With Facebook, you can't do that. I use public relays, but all my notes are backed up and frequently synced to other relays. When people lose their notes, will they perhaps demand a backup/replication service and use more relays? Very probably. And that is what really matters.

Not sure if I understand the graph approach. Do you mean somethingike a "Markov Chain" where redundancy leads to a lower probably of failure? Yes the metrics are an issue. If someone claims that nostr is a decentralized identity layer, I would strongly agree because it gives anyone with entropy an identity. However, I might disagree with claims that it's a decentralized transport layer because we rely heavily on ICANN and IANA to exchange notes.