A vendor claims a conductivity (normalized to 1.0). A lot of adjustable size (twenty by default — a visualization choice, not a sample-size recommendation) is each measured one or more times, with three real-world effects layered in: specimen-to-specimen lot variation, a systematic method–geometry bias, and per-measurement repeatability scatter. With repeats, the tool separates between-specimen spread from within-specimen repeatability — the decomposition a single measurement per specimen cannot perform.
Model: each specimen’s true value = claim × (1 + bias) × (1 + N(0,σlot)); each repeat measurement multiplies in an independent (1 + N(0,σrepeat)); plotted points are specimen means. Between-specimen SD is inferred by subtracting the repeatability contribution of the mean (σ²obs − σ²w/n) — the simplest one-way variance decomposition; real protocols use formal ANOVA-type models and may add fixture and operator levels. The ±10% band is illustrative, not a standard, and the default twenty specimens are a visualization choice, not a sample-size recommendation. Sliders re-use the same fixed lot; the button draws a new one. Percentages are schematic; real budgets are material- and method-specific.