FTA:

  Once PRs start to exceed 10k lines of code, they seem to become slightly “safer.”
  I suspect this is because the extreme end of PR sizes includes refactors, which
  maybe start including less functionality change and therefore have a slightly
  lower chance of breaking. Alternatively, engineers may become progressively more
  reluctant to revert PRs after 10k lines because of emotional anchoring and merge
  conflicts.