Best paper I read was "Conflict-free Replicated Data Types (2011)".<p><a href="https://pages.lip6.fr/Marek.Zawirski/papers/RR-7687.pdf" rel="nofollow">https://pages.lip6.fr/Marek.Zawirski/papers/RR-7687.pdf</a><p>CRDTs get a lot of hype on HN, 95% of the time it's for collaborative editing. But they're much more than some JS library to build an app around - they're a formalism of distributed systems that are strongly eventually consistent. What this means is if the mathematical properties [0] of CRDTs hold, there's no conflicts, no rollbacks, no user intervention - provided the same data is received by every node (in any order, mind you), they will all be in an identical state without a consensus.<p>For me this is massive, and I'm convinced this has big industrial applications, ie distributed systems in domains where the source of truth is most naturally modelled as append only events. In this scenario, the whole database is a single CRDT.<p>Also - and I hope I'm not outing myself as a pleb here - but each time I re-read it I discover new things, stuff I might have glossed over, didn't fully understand, or didn't appreciate before.<p>So yeah, have to hand it to this paper. It's really broadened my horizons.<p>[0] way less scary than you think. If you're comfortable with first year abstract algebra, operations, sets, relations etc you'll be fine.