Adding 313 selected connections removes an adjusted movie response similarity slope in one MICrONS panel

We tested how additions to unrecorded connections can change a link between measured responses and circuit structure. We used public MICrONS v1 data from one mouse. The primary panel had 233797 candidate directions and 6587 recorded connections after quality and variation filters. We removed source-based opportunity strata and additive target effects from response similarity ranks. An exact sorted-score bound showed that 313 selected additions remove the positive linear slope. This is 4.75 percent of recorded connections or 4.54 percent of final connections. Uniform addition of 313 zeros retains 99.86 percent of the slope in expectation. A model feature score needs 446 additions. All four projection groups have positive slopes but thresholds differ. These are sensitivity bounds for fixed measurements and a fixed descriptive model. They do not estimate actual missing connectivity or establish a causal mechanism. One animal, selected candidate directions and unknown error patterns limit transfer. We provide source code, compact data, exact hashes, witnesses, numerical checks and reproduction instructions.

MICrONS · functional connectomics · missing connections · sensitivity analysis · Allen Institute · neural representations · exact bounds