Sensory Fusion Through a Common Coordinate Origin

Sensory fusion is often implemented by synchronising numerical arrays, concatenating features or averaging decisions. These procedures can work, but they do not explain how separately formatted signals become claims about the same persisting world. We tested a more specific architecture derived from a philosophy-of-mind proposal: modality-private systems export relational claims into shared spatial and temporal coordinates centred on the acting system; those claims are attached to continuing entities; and the resulting scene is evaluated against the system's current body boundary before action. A preregistered synthetic experiment required one learned policy to combine complementary visual gesture and auditory intent, identify which visible person produced the sound, preserve body-relative position and event time, and decide whether the current shifted body could use a left or right gate. Across 20 untouched seeds, the unified relational world achieved 76.21% balanced accuracy (95% seed-bootstrap interval 73.63–78.91), compared with 53.18% for late decision averaging, 52.92% for a pre-aligned native package, 65.16% for unbound relational tags, 70.97% for the same specialist-tag package without an explicit binder, and 49.45% when the current body boundary was replaced by its training-population default. The current boundary contributed 26.76 points; explicit binding added 5.24 points over the strongest matched tag control. Corrupting spatial origin, event time or correspondence reduced accuracy, while frozen edits to represented intent and body boundary moved learned action toward the specified counterfactual and sham editing was inert. All 20 preregistered gates passed. The result supports a machine-implementable, sensory-to-asensory world architecture in which the largest contribution comes from organising information around the present self–world relation and a smaller additional contribution comes from explicit entity binding. It establishes computational organisation, not phenomenology.