A Comprehensive Taxonomy of DeFi Attack Patterns: 50 Vectors from 824 Incidents (2017-2026)

We present the first comprehensive taxonomy of 50 distinct DeFi attack vectors, empirically validated against 824 confirmed incidents spanning 2017-2026. Our classification achieves 97.6% coverage (804/824 cases categorized) compared to 58% for the best prior taxonomy. Each pattern includes a canonical real-world example, detection methodology, and Slither detection rule. We find that 8 patterns account for 76% of all losses, with flash loan + oracle manipulation alone responsible for 24% of cases and 60% of total losses ($6B+). The taxonomy reveals that 12 patterns (24%) lack Slither rules entirely, and 18 patterns (36%) require business-logic understanding beyond static analysis. We release the complete taxonomy, detection rules, and an open-source 50-rule DeFi scanner.