Censored by Design: Identifying and Bounding the Level-Cap Bias in Band-Depth Liquidity Measures

Band depth—the notional resting within a fixed percentage band of the mid-price—is a standard liquidity measure in cryptocurrency market microstructure, and is almost always computed from snapshot feeds that return a capped number of price levels per side. We show that this cap right-censors the measure and that the censoring is severe and state-dependent, so band depth is not comparable across regimes. Requesting the same book at several price-grid granularities identifies the censoring multiplier up to a bracket we measure rather than assume: at a one percent band its median across twelve perpetual contracts on one venue is twenty-three, so the default feed captures about four percent of true band depth. Four major venues fail to cover that band at their finest published granularity. Omitted depth is irrecoverable from a capped archive alone, and the evidence is consistent with a conservative bias: crisis-period declines are likely understated.