i have different interpretation that fits the data in the article. the authors found it takes 2 blocks before hyperliquid fills have price information (~700ms) but hyperliquid provides other price information: the orderbook/best-bid/offer and gossip
if best bid/offer rather than fills is the metric for discovering prices, hyperliquid discovers prices every block
we can fill into (1) arbitragers who are trying to pick off stale quotes and (2) non-arbitragers who are willing to execute against the latest quote
hyperliquid batches transaction into blocks, prioritizes cancels/ALO then emits execution results (fills). whereas lighter adds 300ms to all taker orders and executes trades continuously
on hyperliquid, the arbitrage fill signal would be strongest at the first block, and fall off in later blocks
vs lighter, where any signal from arbitragers picking off stale quotes would be at least 300ms after binance and fills from 0-300ms would be dominated by non-arbitrage (2)
if we filtered out arbitrage transactions, i think we would find hyperliquid non-arbitrage fills do discover prices
and hyperliquid gossip should allow latency sensitive users to discover new prices even before transactions are committed
we could view the arrakis article as a measurement of stale quote arbitrage on hyperliquid
the natural followup question would be, does lighter have less stale quote arbitrage?
it wouldn't surprise me if lighter has less arbitrage because they continuously execute trades which minimizes jitter
the reason for this could be that hyperliquid executes and prioritizes cancels in blocks: even with cancel priority, if the price moves close to block a boundary an arbitrager can still land first, especially with write priority
the counter-factual would be we see a spike of lighter correlation to binance at 300ms which i dont see in the article