KL Divergence
D_KL(P||Q) — how one distribution diverges from a reference
KL divergence (Kullback–Leibler; Kullback & Leibler, 1951) measures how much probability distribution P diverges from reference Q. Not a metric (asymmetric), but standard for observation quality:
Quality(sensor) ≈ 1 − D_KL(P(observation) ‖ P(ground_truth))
A calibrated, traceable instrument → near-zero divergence. An uncharacterised sensor → unknown divergence — and an ceiling on trust for everything derived from it. Estimating divergence from finite calibration data is itself uncertain (Bayesian hierarchy; pragmatic tier-one boundary).
Corpus stance
A2 — working context: Formalises “characterise and publish observation quality”; see What Information Theory Says.