Transactions in LN are just as "reliable", and don't introduce centralization in any meaningful way.

As for the "modeling", I'm not sure what you want. The threats have been outlined in the whitepaper, and successfully tested in some capacity on testnet. And now they are being tested as lightning network is being rolled out on mainnet. There might be some formal "threat modeling", but i'm not familiar with what that would even look like or mean.

No better way to "threat model" than to try it out in a hostile environment where bad actors that have some kind of "exploit" can already use it to gain BTC.