With positive linear transfer and count weights, motor input kept 62 of 65 local muscle pull directions within 15 degrees in one Platynereis larva

Can shared motor input preserve the local forces that muscles could make on their own? We joined neural and muscle attachment maps from one three-day-old Platynereis dumerilii larva. A straight tensile model gave 148 geometry-qualified muscle-to-rod pairs. Only 65 pairs, from 64 muscles and 68 motor neurons, had retained input in the filtered neural graph. With positive linear transfer, count weights, equal muscle gains and no activity limits, shared input came within 15 degrees of 62 of 65 independent-muscle pull directions. Allowing one gain per muscle from 0.5 to 2 raised this to 64 of 65. The largest error fell from 34.82 to 26.98 degrees. It remained 27.38 degrees in the second attachment export, but fell to 13.48 degrees with binary input weights. Full linear rank did not ensure access with positive activity. The test concerns local force direction with other rods unconstrained. It does not measure force or movement. A full-body follow-up failed numerical checks and supports no global claim. The package gives the new force atlas, a convex test for bounded muscle gains, all 912 local solutions, source hashes, figures, independent checks and offline repeat steps.

connectome · motor control · muscle attachments · Platynereis · force direction · positive linear model · muscle gain · reproducible analysis