A marimekko reaches plotly as stacked bar traces whose
columns are drawn at precomputed cumulative positions and given a
per-column width. MAIDR does not guess at
that: width is ordinary bar styling that any bar trace may
carry, and reading a non-uniform array as data would announce every
column's width as a share of all observations — a number the chart does
not contain.
So a mosaic is declared, on plotly's own metadata attribute:
meta: { maidr: { type: 'mosaic' } }. With it, each cell
announces its column's share of all observations alongside its own
proportion, and the description names the largest and smallest group —
the marginal distribution a reader cannot accumulate by walking the
segments. width and count may name columns of
customdata when the figure carries the contingency table it
was drawn from; without them the drawn widths are normalised into shares.