SQA::BI predicts a probability distribution over a small ordered set of outcomes (for example strong downtrend through strong uptrend) from a numeric feature vector. Priors with Laplace smoothing, Gaussian kernel density likelihoods, and a posterior that reports entropy, confidence and KL divergence from the prior. An LLM can optionally supply the prior, or act as a likelihood function over textual evidence — the LLM only ever judges, Ruby does all the probability arithmetic.