Maximal Extractable Value (MEV) on high-throughput blockchains can be captured through targeted search, where bots identify opportunities off-chain and submit route-committed transactions, or through probabilistic search, where bots submit repeated attempts that resolve opportunity discovery during on-chain execution. This distinction has direct implications for spam, blockspace consumption, and protocol fee revenue. We model how ordering granularity, fee floors, and opportunity-access shocks shape competition between these architectures. Using cyclic arbitrage data on Base from June 2025 to February 2026, we develop a trace-level classifier for search architectures and show that the resulting labels correspond to distinct execution behavior. We test the model across three episodes: Flashblocks selects against broad on-chain probabilistic scanners; token-launch opportunity shocks temporarily revive probabilistic search; and higher fee floors select against probabilistic bots whose opportunity flow cannot sustain repeated attempts. In our sample, probabilistic search accounts for only 23% of arbitrage activity but produces 95% of spam and consumes 20% of Base gas. After Base's configuration changes, protocol fee revenue shifts toward successful arbitrages and away from spam, probabilistic bots pay higher priority fees, and spam consumes a smaller share of blockspace.

翻译：高吞吐量区块链上的最大可提取价值（MEV）可通过定向搜索（机器人离线识别机会并提交路线承诺交易）或概率搜索（机器人提交重复尝试，在链上执行期间解决机会发现）来捕获。这一区别对垃圾交易、区块空间消耗和协议手续费收入有直接影响。我们建模了排序粒度、费用下限和机会获取冲击如何塑造这些架构之间的竞争。利用2025年6月至2026年2月期间Base链上的循环套利数据，我们开发了一种针对搜索架构的交易层级分类器，并表明由此产生的标签对应不同的执行行为。我们在三个场景中测试了该模型：Flashblocks选择性地排斥广泛的链上概率扫描器；代币发行的机会冲击暂时复兴了概率搜索；更高的费用下限会淘汰那些机会流无法持续支持重复尝试的概率机器人。在我们的样本中，概率搜索仅占套利活动的23%，却产生了95%的垃圾交易并消耗了Base链20%的Gas。在Base链配置变更后，协议手续费收入转向成功套利而非垃圾交易，概率机器人支付更高的优先费用，且垃圾交易占用的区块空间份额有所下降。