A perp or spot fill on-chain is not one timestamp. It is block time, oracle time, and liquidation time. Desks that treat those as the same clock misprice risk.
Central-limit books on a matching engine share a single clock: the exchange’s sequence. On-chain trading does not. The block (or sequencer slot) that includes your transaction is one clock. The oracle that marks the contract is another. The liquidation engine that may take the position is a third. They are allowed to disagree, and they often do.
That disagreement is not a bug in the UI. It is the product. A maker on an AMM sees a reserve ratio. A taker on a perpetual CLOB sees an index and a mark. A liquidator sees a maintenance threshold against an oracle that may update on a slower cadence than the book. If you design infrastructure as if those three numbers were one ‘price,’ you will ship a pretty terminal and a broken risk engine.
Professional on-chain trading therefore starts with latency as a cost, not as a slogan. Inclusion delay, oracle heartbeat, and funding interval are the same class of object: they determine whether a hedge is still a hedge when the next block lands. Fee floors, blob markets, and L2 sequencing rules change those costs without changing the user’s intended trade.
This desk does not publish live quotes or signals. Selected work is where we show protocol and product studies. This blog is where we write about the market itself — execution, structure, risk, and MEV — in the same register as the journals the field already trusts.
If you only remember one operational rule: never let a portfolio view imply that mark, index, and last trade are interchangeable. Label the clock. Operators who cannot name which clock they are looking at are not trading. They are decorating a number.
Desk notes on on-chain trading. Not investment, legal, or tax advice; not an offer of securities; not a live quote or a signal. Protocol and product studies live under Selected work.
← All essays