This prints a "compatibility" result, for a certain definition of
"compatibility":

*   we look at the results in order, and consider we failed at the
    first result which doesn't match;
*   for every matching result we get points;
*   the first page is worth more point than the second, which is worth
    more points than the third, which is worth more points than the
    fourth, which is worth more points than the fifth; all pages after
    that are worth the same amount of points as the fifth;

Hopefully this is not a terrible definition of "compatibility", as it
favours the first pages of results, which really are what most people
care about. Being wrong in the later pages is still counted as
incompatibility, but it's not as bad as being wrong at the start.

The compatibility target (threshold) can be passed as an argument, so
that we can eventually run this in a CI and try to improve our
compatibility.