Plate difficulty estimates the expected play count to clear a plate, adjusted from the FC/SSS/AP achievement rate in the selected rating bucket.

It mainly reflects how hard the charts are, and less how many charts the plate contains.

Boring, pedantic math notes

First, raise the data point $x_i$ to the $\alpha$ power to estimate the per-play FC/SSS/AP probability ($\alpha$ is $1.4$ for SSS/FC/AP plates and a steeper $1.5$ for FSD, which needs both players in a matching session to clear ≥ FC+ at once). The aggregated data is not the success rate of a single play, but the share computed from each user's best record in the rating bucket. Take the reciprocal to estimate the play count $c_i$. To downweight how much the chart count contributes to the total difficulty, we then subtract $1$.

Lower-level charts also tend to inflate the play count for players who simply haven't played them recently, so we multiply by a per-level correction:

Finally, the per-chart values are summed across the plate; we shift the result so the final difficulty roughly falls between $0$ and $10$ for readability: