Implement the sim-core WorkGrid shape for one-mode machine delegation, wired demand/knowledge routing, physical exposure absorption, and queue snapshots as the render contract. Mark machine-work in progress so follow-up Sim/save/frontend work builds against this substrate instead of the older per-machine split model. Defense: This is the smallest executable slice of the machine-work spec: the constitution says work is somewhere and the flow law says tokens ride one substrate. Landing the substrate first prevents frontend counters or split allocation UI from becoming a second truth while the rest of the machine-work loop is wired.