Fog-of-war for any domain: reuse one generic beliefs table + a secrecy blend

To add perceived-vs-actual (agents decide on beliefs, not ground truth) to a new domain in a sim, do NOT build a parallel belief store. If a keyed (holder_id, topic, key) beliefs table already exists (e.g. for market prices), add new topics (pol:power / pol:holdings / pol:rivals) to the SAME table and you inherit its confidence-decay and aging for free. Make the score-facing view function observer-relative: view(store, observer_id, target). Self returns truth; unseen targets return a PRIOR (not truth); seen targets return the last sighting, aged. Crucial: make belief diverge from truth via a single secrecy blend at observation time: believed = true*(1-s) + projected*s, where s in [0,1] is the target guardedness and projected is what it broadcasts. That one line gives you bluffing (weak actor, high projected, high s) AND hiding (strong actor, low projected, high s) as emergent, authored-once behavior. Gate the whole thing behind an opt-in flag so default runs are byte-identical (regression-safe).