Human loop — Orient → Judge → Steer → Learn → Refine.

The recurring user practice around the machine Program, not a second execution pipeline. Orient means understanding material change relative to the maintained model; Judge means supplying the smallest useful human decision at an escalation boundary; Steer means pointing the read at a question that matters and feeding it sources or corrections (skin in the game — the user owns the read); Learn means following the read's re-prices, their reasoning, and the evidence trail; Refine means preserving the correction — and how the read moved — so later runs start sharper. The human need not perform all five on every cycle. Most runs require only orientation; judgment appears when consequence or uncertainty earns it. See Product behaviours.