This is the first optimistic rollup I’ve seen that uses ZK where it actually hurts: fraud, not vanity. @megaeth_labs is wiring non-interactive ZK fraud proofs into the dispute path so bad state gets nuked in one shot no griefing games, no days-long bisection tennis. ❯ What I dug into: Mega’s ENDGAME write-up (Jul 27, 2025) spells it out challenger generates a single proof in the RISC Zero zkVM, following the OP Kailua hybrid model; the verifier on L1 accepts/rejects in one tx. ❯ Why it matters: you still keep the standard 7-day detection window, but once a challenge lands the dispute settles in ~1 hour instead of dragging finality for days. ❯ How it stays honest: pairing with EigenDA means watchers can always reconstruct the block payload to prove fraud; no withhold-and-stall escape hatch. ❯ Cross-checks: RISC Zero’s Kailua posts describe the same non-interactive ZK fault-proof design for OP chains fast finality without turning the whole thing into a full ZK rollup; Boundless/Conduit document the operational path for production OP-stack chains --- My read: this keeps the optimistic economics (cheap, simple) while deleting the worst grief vectors. The open question is prover cost at scale (Mega’s own estimate puts worst-case around $100/dispute in cycles/$$ terms) totally fine if fraud is rare, but worth monitoring as usage explodes. If you’re building, assume “final if unchallenged; fast if challenged.” That’s a nicer contract to code against than “final… eventually.”