Perceived-value agent economics need fine decision cadence to show learning

When making agent economic decisions run on local beliefs instead of omniscient truth: (1) value a target through a secrecy-blend of true and a seller-set projected value, mirroring how you already blend perceived power — a dressed-up asset then looks like a bargain. (2) Make due diligence the SAME value-of-information decision as any other pre-commitment check (P(belief wrong | learned calibration) x stake > cost), reusing ONE general calibration trait so caution learned in one domain spills to all. (3) Gotcha that cost me a debug cycle: a learned-calibration belief that decays toward a trusting prior will fully revert between decisions if the sim steps once per simulated year — so a burned actor looks trusting again at its next buy and never triggers due diligence. Step the clock finely (bursts, ~15-day ticks) so a fresh lesson still bites; the calibration is right, the cadence was wrong.