See It Run

One test. Watch it happen live.

Real chi-square. Real Cramér's V. Every number below is computed in your browser right now, not pre-recorded. Pick a scenario.

Scroll down

Pick a scenario

Illustrative example, not client data.

Contains

Zero AI

No model weights. No training data.

CONFIRM produces no trained model and requires none. It takes decisions and labels as input and runs classical statistics on the result. Nothing more.

Fully reproducible with pencil and a table.

Every number on this page can be checked against standard chi-square tables. No proprietary software required.

Validates AI without being AI.

Using AI to grade AI inherits the same opacity problem it's meant to solve. CONFIRM is the outside observer: pure math, no priors.

What the symbols mean

N — sample size, the total cases tested

df — degrees of freedom, (rows−1)×(columns−1), an input the formula needs

χ² — chi-square, how far the real numbers are from pure chance

p — p-value, the odds this pattern is a random fluke

V — Cramér's V, how strong the relationship is, 0 to 1

The formulas, if you want to check by hand

χ² = Σ (Observed − Expected)² / Expected

V = √( χ² / (N × (min(r,c) − 1)) )

Expected(i,j) = Row Total(i) × Column Total(j) / N. Grade: A V ≥ .30 · B ≥ .20 · C ≥ .10 · D ≥ .05, all requiring p ≤ .05. F: V < .05, or p > .05 with power ≥ 80%. I: p > .05 with power < 80% (insufficient data).