This is exactly my point. They listed the top-line, total number of all PRs without any filtering, then when they calculated the "spam ratio" they filtered out "invalid repos" as well as "spam PRs". It's a misleading statistic.