Every Jumper Against Every Other

At a long-jump trial, each athlete's best jump is recorded in whole centimetres. The coach wants every athlete compared with every other one in a single grid, where

is how much further athlete $i$ jumped than athlete $j$ — negative if $i$ jumped shorter. Row $i$ is athlete $i$'s view of the whole field.

Jumps only a few centimetres apart get argued over, so athlete $i$ only clearly beats athlete $j$ when $\text{gaps}[i][j]$ is strictly greater than margin.

Task: write head_to_head(jumps, margin) returning a tuple (gaps, clear_wins):

• gaps — the whole grid as a list of lists: one row per athlete and one column per athlete, both in the order given;
• clear_wins — for each athlete, in order, how many others they clearly beat.

margin is never negative.

Example

Athlete 0 out-jumped athlete 1 by 25 cm — more than 20, so it counts — but fell 28 cm short of athlete 2. Athlete 2 clearly beat both of the others.

A grid of every sample against every other sample is the starting point whenever you ask which examples sit closest together. This is the one-feature version.