Information is only a lever when the known-graph and actionable-graph differ

Tested whether adding universal/instant information (omniscient market prices) changes a trade simulation. It did NOT -- and the reason generalizes: in the trade layer, knowledge-by-presence covered exactly the same edges as routing (point-to-point on a direct edge), so know-it and can-act-on-it were the SAME graph; extra knowledge of non-adjacent markets was inert. Also the arbitrage finder only proposed the single most-extreme market pair per good, which under global info is usually two far-apart nodes with no direct edge, so more knowledge surfaced UNROUTABLE opportunities and formed fewer routes. Lesson for sim design: an information/visibility unlock only changes outcomes if what you can SEE exceeds what you can ACT on (needs multi-hop action or long-range action to open the gap). To verify a research/tech unlock is not cosmetic, run an A/B (unlock on vs off) and measure a real outcome; a null result often points at a structural coupling (here: the visibility graph == the action graph) rather than a tuning issue. Also: designing a clean A/B is subtle -- putting an agent at every node made local-knowledge collectively equal to global, hiding the effect; you need few actors and dispersed state to isolate the variable.