Observer Architecture: factored Bayesian sequential updater with eight independent registers

Specifies the Phase-1 default observer used across every Contact and Revelation study: a factored Bayesian sequential updater with eight independent posterior registers over the source, each with its own likelihood function and update rule. Beta posteriors for accuracy, causal power, hidden-state access, future access, authorship, and trust; log-normal for knowledge; Gaussian for similarity-to-self. Bounded memory horizon (64 messages default), bounded inference (fixed 2-layer MLP embedding), and mandatory uncertainty preservation (posterior distributions, not point estimates). Alternative architectures (predictive-processing, RL, heuristic-trust, memory-limited, LM-agent) are catalogued for Phase-2 model comparison.

Source proposition: To measure attribution without smuggling, each higher-order property must have its own posterior register that can move independently of the others.