Farcaster is a fine demonstration in why you don't need strict global consensus and public blockchains for most things, but at the same time, they can prove valuable for 1 or 2 things

I hope to see more hybrid applications that are significantly better by minimizing blockchain usage only to strict global consensus

For vertical use cases we can maybe think about this as enforcing compute rules (see as spectrum with quality ratings?) at protocol level - biz as usual today, & cheaper w/ time - to determine how your db shd end up with incentivisation for enrichment. Rly depends how "rich" data for said use case shd be perhaps.