Chainlink: the candle's close block is drawn by Chainlink VRF 2.5 through a dedicated provider contract, live on Base Sepolia with multiple epochs settled by real fulfillments on a funded subscription. Unichain: the primary interactive demo runs on Unichain Sepolia, where an autonomous keeper drives the full candle lifecycle in about thirty seconds; since no VRF grade randomness service serves Unichain Sepolia yet, that deployment uses a clearly labeled blockhash provider behind the same pluggable interface, ready to swap the moment one arrives.
Github: https://github.com/big14way/wick Slides: Project Link: https://wickhook.vercel.app Demo Video: https://youtu.be/StR6x8n5xDE
Sandwich bots extract hundreds of millions of dollars a year from AMM swappers, and every existing defense leans on a crutch: private orderflow trusts a relay, commit reveal costs a second transaction, and batch auctions with a fixed close just move the fight to the last block, because a known deadline is a race that MEV bots always win. The inspiration came from seventeenth century England, where candle auctions kept bidding open while a candle burned and nobody knew when the flame would die, a trick Polkadot later revived for parachain auctions. Wick brings that mechanism to swap settlement: orders accumulate during a window, a random close block inside it is drawn retroactively, and every order at or before the close settles at one uniform price per side. You cannot position around a deadline that did not exist when your transaction landed.
Against the official dataset of all 715 prior hookathon submissions there are zero candle auctions, zero uniform clearing hooks and zero probabilistic closes; Wick's protection is structural rather than statistical. The test suite proves it adversarially: the same sandwich attack that clears +7.87 tokens against a vanilla pool finishes down 0.33 against Wick on identical liquidity, because custody gives the attacker no price movement to straddle and uniform clearing gives them no victim slice to carve out. On the sustainable liquidity side, the instant lane converts MEV pressure into LP yield, a volatility priced fee paid entirely to liquidity providers, so pools keep both patient and impatient volume instead of losing users to private relays. Everything is live and reproducible: 49 passing tests with 98.77 percent line coverage including fuzz and invariant campaigns, verified contracts on two testnets, a public dashboard, and an autonomous keeper settling epochs in about thirty seconds.
The hardest lessons came from live testing rather than theory. A tester blowing through the book pinned the pool at the minimum tick and revealed that settlement could revert while price sat beyond the deviation bound, and that a volatility spike could pin the instant fee at its cap; both became contract fixes (settlement now always completes with the residual refunded, and the fee is hard capped with time decay), new adversarial tests, and a redeploy of both chains. Getting custody right inside v4's flash accounting, pairing the BeforeSwapDelta NoOp with ERC6909 claims so every wei stays provably backed through netting, bounded residual swaps and pro rata refunds, took an invariant suite to trust. And discovering that no VRF grade randomness serves Unichain Sepolia forced the pluggable provider design, which ended up strengthening the project: the same hook demonstrably runs on demo randomness and production Chainlink VRF without changing a line.