Every quantity is computed from the standard formulas, with the conventions statisticians actually argue about made explicit. Quartiles and percentiles use R’s default "type 7" linear interpolation: h = (n − 1)·p + 1 on the sorted sample, so Q1/Q3 match what R, pandas and Excel’s PERCENTILE.INC produce. Variance and standard deviation are reported in both flavours — sample (divide by n − 1, the unbiased estimator for a sample) and population (divide by n) — and the variance-basis switch decides which one drives the coefficient of variation and the standard error of the mean. Shape statistics follow the moment conventions: skewness comes both as the biased moment ratio g1 = m3/m2^(3/2) and as the adjusted Fisher–Pearson G1 = √(n(n−1))/(n−2)·g1 that Excel’s SKEW returns; kurtosis is excess (normal = 0), both biased g2 = m4/m2² − 3 and the small-sample-adjusted G2. Outliers use Tukey’s fences (Q1 − 1.5·IQR, Q3 + 1.5·IQR) from the same type-7 quartiles. When you paste a second column, the calculator adds the bivariate suite: sample covariance, Pearson’s r on the raw pairs, Spearman’s rho computed as Pearson on average-assigned ranks (the tie-correct definition — the 6Σd² shortcut is wrong once ties exist), and the least-squares line ŷ = slope·x + intercept with R². Input accepts any mix of spaces, commas, semicolons and tabs, plus scientific notation; tokens that are not valid numbers are rejected up front rather than silently dropped.