Emergent NPC caution: general calibration + scored VoI, not a named flag

Self-review caught a scripted behavior smuggled into an emergent sim: I had a belief literally named wisdom:scout, pumped only by repelled raids, gating scouting via a threshold — a rule named after its outcome. The emergent refactor, three moves: (1) Replace the purpose-named accumulator with ONE general trait — appearance_reliability (how much outcomes have matched what things looked like). (2) Update it by prediction error whenever ANY belief is TESTED against reality, both directions (a raid confirms or refutes I-can-beat-this: success raises trust, a repel lowers it) — not a hand-wired event->disposition link. (3) Make the action a SCORED decision, value_of_intel = P(wrong|calibration) * stake vs the action cost, so scouting FALLS OUT instead of being triggered by a flag. Now the trait is general (same calibration gates an acquisition or a pact), scales with stakes, spreads by diffusion, fades by revert-to-prior. Test for enforcement vs emergence: does the fix DELETE authored content (a named belief + a threshold) and let behavior emerge from a value computed off real state? If yes its emergent; if you added a term named after the outcome, its forcing.