Plotly has no word-cloud trace: the documented recipe is a
scatter in text mode whose textfont.size is an
array, one glyph size per term. That per-term size is what MAIDR detects.
The layout of a cloud carries nothing — the coordinates are chosen to pack
glyphs, not to encode anything — so the terms are walked in weight order
instead, and the weight itself is taken from customdata where
the author put the real counts there, as below, rather than from the glyph
size that stands in for them.