Maybe significantly more work, but can you try to filter out very small (or spam) PRs before these calculations? Maybe limit this analysis to PRs that edits sou

Maybe significantly more work, but can you try to filter out very small (or spam) PRs before these calculations? Maybe limit this analysis to PRs that edits source code (not just comment and docs). Or maybe include those, but only when "amount" of changes cross a certain threshold? (e.g. exclude PRs that fix just one typo)