Two live-sim gotchas: wide sight is not deep planning; agents pollute their own price signal

Debugging a multi-hop trade sim surfaced two generalizable bugs. (1) Giving agents WIDE information (they can see all markets) is not enough to get lookahead behaviour — planning DEPTH is a separate knob. A merchant that saw all 8 markets on a chain but had a 1-hop planning horizon stayed myopic and opened zero relay routes; the reservation/value field correctly showed the far value routed back, but the DECISION used a shallow lookahead. Fix: raise the planning-horizon knob (here an info_regime flag), not just belief coverage. Lesson: see-vs-plan are orthogonal; a myope wastes a data-grid. (2) When market price is derived from supply-at-a-location AND that supply counts entities located there, an agent that HOLDS inventory at a market inflates that markets supply and flattens the very price gradient it trades on — a feedback loop that erased the scarcity signal (dear end supply 12 became 72 with the merchants own 60 units, price 46 -> 8). Fix: decouple the demand/price signal from the agents own stock (exogenous demand, or exclude the actors own holdings from the market pool). Both hid a working mechanism behind a config/feedback issue, not a logic error.