Where Monad actually trades: Kuru, Uniswap, and the rest of the book

A chain is not a market. The market is whichever contracts currently hold the inventory. On Monad in October 2026 that inventory sits in two kinds of place. One is the set of Ethereum protocols that deployed bytecode they already trusted: Uniswap, PancakeSwap, Curve, and the aggregators that sit on top of them. The other is native venues built for the block time, of which Kuru is the one the volume numbers keep returning to.

CertiK’s note of 20 October 2026, using figures through the middle of the month, said Monad DEXs had done about $2.64 billion in thirty days. Kuru accounted for about $1.25 billion, which is 47 percent. An August record day was reported at $362 million inside a month of $3.64 billion. Treat both as a snapshot from one researcher. They are large enough that “illiquid testnet” is no longer a serious description, and concentrated enough that a routing algorithm which ignores Kuru is leaving half the printed volume on the table.

Order book, not just a curve

Kuru is not only an automated market maker. It runs an onchain order book and, through 2026, rebuilt parts of that book around changes in how Monad stores data. Its builders have talked about a block gas limit of 150 million and a replacement rate, if the whole block were given over to cancels and replaces, in the tens of thousands a second. That is a ceiling, not a Tuesday. The more interesting published figure is operational: a reconciliation pipeline they put at 115 microseconds at the median and under 690 microseconds at the 95th percentile, per block. The point of those numbers is not the brag. It is that a 300 millisecond block makes the indexer part of the exchange. If your view of the book is a second old, you are three blocks late.

Half the printed volume in one venue is a fact about routing. It is not a fact about safety.

How to read a pool before you touch it

Volume is not depth. A pool can print a large number because the same inventory turned over, and still move badly when you arrive with size. The checks that matter are the ones a router can do in the same block you intend to trade: output quoted, fee tier, and whether a second independent node agrees the quote. MONAF’s own desk refuses a round trip that does not clear its cost against the target you set. That is a local rule, not a property of Monad. The chain will execute a bad swap just as quickly as a good one.

A book has to live inside the slot

An automated market maker keeps two reserves and a curve. An order book keeps every order that has not been filled or killed. The second design is only a market on a 300 millisecond chain if a cancel arrives in the slot the price still refers to. Kuru has published a reconciliation pipeline at 115 microseconds in the median and under 690 at the 95th percentile, per block. Read that as a claim about one venue’s machinery. Then time your own screen. A quote that is one second stale is three blocks stale, and the book you are about to lift may already have been lifted.

Half the tape is a single venue

A router that never asks Kuru is, on CertiK’s mid-October cut, ignoring 47 percent of printed DEX volume. The converse is just as sharp. If that venue’s matching stops, half the tape stops printing, and Uniswap, PancakeSwap and Curve do not automatically inherit the depth. They are separate contracts with separate inventory. Familiar bytecode is not a mirror of the native book. Concentration is a fact about where to route. It is also a fact about what breaks when one matching engine has a bad hour.