Finer numerical grids barely change average error, yet rounding can change a threshold decision; statistical undercoverage remains. The synthetic TP-01 study has 24 populations of 128 participants per regime, each with 40 recorded observations. A–B show mean absolute error and coverage of nominal 90% intervals; whiskers are 95% t intervals across populations. C shows 24 paired grid differences per regime (96 total), deduplicated 199-point minus conventional 1,001-point, with mean diamonds and 95% t intervals. D uses filled circles for continuous Beta lower 5% quantiles and open squares for grid values at .75; exact gates return REVIEW. Known-ID deduplication improves coverage but increases point error under the fixed prior. D uses a separate reference-risk/no-decay profile; neither clinical calibration nor estimator superiority follows.
