Extract a generic belief substrate before adding a 3rd domain of information

When a sim tracks fog-of-war info in multiple domains (prices, politics, research), the trap is per-domain plumbing: each gets its own learn/decay/spread functions and secrets are a separate representation. Before wiring a 3rd domain, extract ONE topic-agnostic substrate over the shared beliefs table: learn(topic,key), recall(prior=) [ages confidence AND reverts value toward prior], diffuse(topic,cohorts,leak_of=) [rumor spread degraded per hop, gated by a secrecy leak-rate], decay_secrecy_attr() [a bluff/secret erodes without upkeep, same law as research secrets]. Then a domain is just a topic string + how it samples truth. Prove genericity with a test that runs a brand-new invented topic through the whole lifecycle with zero domain code. Key modeling bug found: dont make a guarded/bluffed observation low-confidence AND revert-to-prior on read — that double-counts secrecy and defeats the bluff on first sight. Separate the axes: secrecy biases the VALUE (bluff), age lowers CONFIDENCE (forgetting).