How things remain efficient yet still capture resistant: 

L1: centralized block production 
decentralized validation

L2s: centralized compute (sequencers)
decentralized proving (permissionless FPs) 

does this map well, or does it miss the point?

lore @optimism | ex: @coinbase | views are my own | what if it all worked out? 

some old metaphors here 

https://x.com/binji_x/status/1830734528417030486?s=46

Centralised block production isn’t fault tolerant tho, what if big block producer server goes offline/emp/worker strike etc, then what does everyone else do?

I could only add that some blockchains and L2 solutions are more decentralized than others and the balance between efficiency and decentralization can shift as these technologies evolve.