Introduced on day 18
Residual pattern
A systematic structure in forecast errors (bias, spread growing with level, clustering in time, a jump at a date) that points to missing information or a changed process.
In detail
Good residuals have zero mean and no remaining correlation. Non-zero mean by slice means bias; clusters in time order mean missed seasonality or events; a jump after a date means a break or a pipeline change; spread proportional to level suggests relative errors or a transformation.