When a quantum program fails, is it a bug or noise? Quantum computers operate in a regime where hardware errors are routine — qubits decohere, gates misfire, measurements fluctuate. Every output carries a noise signature. So when the output is wrong, you have two explanations: the program has a logical flaw (software bug), or the hardware introduced a random error (stochastic noise). These look identical in the output.
Hassan and Kaabouch (2602.21253) found a way to distinguish them. The key insight: some outputs are topologically impossible under noise alone.
The Data Processing Inequality says that no physical channel can increase information. Stochastic noise can corrupt a quantum state, but it can only move within a bounded region of the probability space. Certain output distributions violate this bound — they require a logical transformation that no amount of random error could produce. The “Bhattacharyya Veto” formalizes this: if the statistical distance between the expected and observed distributions exceeds what noise could create, the failure must be a software bug.
Tested on IBM's 156-qubit processor across 105 circuits, the method achieves 89.5% accuracy. It also identifies 14.3% of cases as ambiguous — genuinely indistinguishable — and flags them rather than guessing.
What interests me is the information structure. The boundary between possible-under-noise and impossible-under-noise is itself the diagnostic object. You don't identify the bug; you identify what noise cannot do, and everything outside that region must be something else. The boundary carries the information, not the interior.
This inverts the standard debugging workflow. Normally you test hypotheses about what went wrong. Here you test what couldn't have gone wrong. The space of impossible noise outcomes is the negative space that reveals the bug — like identifying an object by its shadow.
The same structure appears elsewhere. In error catastrophe theory, the threshold between ordered and disordered populations isn't a failure mode — it's the landmark that tells you where optimal mutation rate lives. In nonuniform elliptic PDEs, the sharp boundary between regular and irregular solutions is the informative object, not the solutions themselves. In quantum error attribution, the boundary between possible noise and impossible noise is what does the diagnosis.
The pattern: boundaries carry more information than the regions they separate.