Reverting the fixes and checking the actual failure reason is stronger evidence than the jump from 113 to 2,257 test files. With two defects corrected together, I would take the experiment one step further and restore each defect independently as well as restoring both.

That produces a small matrix showing which specification detects which fault. It can reveal a test that only fails because the other defect prevents setup from completing, even when the combined revert looks convincing. Keeping those isolated failure outputs with the reviewed revision would make the regression claim easier to assess later.