Fusion Beyond the Buzzword: Reviewing RF and Acoustic Evidence Together

Frid et al., Sensors (2024), evaluate RF and acoustic evidence using selected datasets and report improved classification results, particularly in their noisy evaluations. Those evaluations include added Gaussian noise: a defined experimental condition, not every disturbance a deployed system might encounter. The defensible claim is that combining these inputs helped within the reported study. Recasting that result as universal noise immunity would erase the condition that makes it meaningful. Neither the paper's publication nor its result establishes performance for a particular commercial integration.

Agreement between modalities is not automatically independent corroboration. Both may be evaluated against the same narrow collection of circumstances, leaving shared gaps invisible. An omitted background condition, for example, can undermine conclusions drawn from otherwise encouraging results across both inputs. Correlated errors mean failures occur together more often than an independence assumption would suggest. This is a conceptual caution, not a failure established by the Sensors paper. Evidence of complementary inputs should not be presented as evidence that common failure causes have disappeared.

CISA's guidance (2025) asks buyers to examine false positives, known limitations, testing and test environments. For a fusion proposal, the relevant evidence should explain whether the comparison concerns the offered product or only an external research result. A report should make the evaluated product version and environmental scope clear, alongside limitations and error outcomes. Compatibility documentation establishes that components can work together; it does not establish an accuracy benefit. A missing product-specific evaluation leaves that benefit unverified, not disproven.