Verify a claim the way a fact-checking desk does. Up to six independent-looking checks (reverse image search, source reputation, corroboration, metadata, account pattern, fact-checker database) each give a noisy score; the lab adds their log-likelihood ratios to the prior log-odds on a balance scale and flags the claim when the posterior passes your threshold. Raise the shared-origin correlation and see why counting copied reports as separate evidence makes naive Bayes overconfident, while correct fusion stays calibrated. Blender-built scale, gauges, news card and magnifier, with closed-form and Monte Carlo checks. Synthetic educational model.