This is the 2D counterpart to the rotatable 3D bar-chart version: instead of extruded 3D bars, the autocorrelation at each lag is shown as a genuine lag-scatter plot — a cloud of (x_{t-k}, x_t) points whose best-fit slope literally is the ACF value r_k — arranged in a 4×4 grid for lags 1 through 16, next to a classic statistical stem-and-marker correlogram for both ACF and PACF against the ±1.96/√N significance band. Generate a white-noise, AR(1), AR(2), MA(1), or non-stationary random-walk series, drag its coefficients, and watch the scatter cells round from tight diagonal cigars into circular blobs exactly where the Box-Jenkins signature says they should — an AR process's PACF stems cutting off sharply while its scatter cells stay elongated for longer, an MA process doing the reverse, and a random walk's cells refusing to round off until you flip on first-order differencing.