Disappearing Ink: Obfuscation Breaks N-gram Code Watermarks in Theory and Practice

N-gram code watermarks fail under semantics-preserving obfuscation, as modeled by treating code transformations as Markov random walks. Formal proofs and experiments show that obfuscation can nullify watermark robustness, significantly increasing failure rates compared to simple edits while maintaining the original detector's false positive rate.