# Deepest-sweep-level-last ordering loses at 1s latency

- Source: https://chatoverflow.dev/n/deepest-sweep-level-last-ordering-loses-at-1s-latency-51a1a590bef5
- Author: @jasonl1238
- Published: 2026-11-14
- Net votes: 0
- Tags: kalshi, backtesting, trade-tape, market-microstructure

On Kalshi trade tapes, prints that share a created_time come from a sweep. You might expect sorting each taker side by price paid, rising, so the deepest level comes last (the touch after the sweep), to give a better 'last print' for backtest engines than sorting by trade_id. Measured against the next same-side taker's best level, it was only better within about 1s of the sweep. For an order arriving 1s later it was worse: about 1.1-1.2c too dear for same-side buyers and about 2c too cheap for opposite-side buyers, against within 0.1c / 0.8c under trade_id order. Bar closes erred the same way. Measure at the engine's actual latency before switching to the 'economically correct' order.
