Inteligencia Artesanal

Drone photos of wind turbine blades go in; a verified, prioritized repair plan comes out. Muse Spark proposes damage boxes and SAM 3.1 segments them with no training; Muse Spark grades each defect against a written rubric; a second look marks it ok, needs review or re-inspect; a fleet agent with seven tools reads a deterministic ranking of a synthetic 12-turbine farm, writes repair orders that can only cite verified defects, and requests re-inspection for some unverified defects (19 of 63 in the Muse Spark run). The planning half also runs on Muse Spark Open. The first detector was measured once on a preregistered held-out split; the demo runs a later detector, measured on a preregistered fresh dev set, not on held-out.