From popularity metrics to probabilistic confidence and uncertainty
V2.0.0 introduces the core insight: credibility should be modeled as belief under uncertainty, not popularity. The system does not attempt to measure truth directly—it estimates conservative confidence given noisy evidence.
Established the initial mapping from user behavior signals to belief updates (positive vs negative evidence).
Confidence is not hidden. Low-evidence items remain uncertain rather than being ranked as “bad.”
Established initial consistency checks and safe defaults for missing data (priors + conservative fallbacks).