A red-team run seeds small bugs into a file and counts how many its tests catch. Before it could start on the prop-firm challenge calculator, the baseline was already red: the page test treated the inline data: favicon as a file link and failed on "broken internal link". Four sibling ventures shared the regex and one more (portable-restroom) tripped the same favicon in an external-reference check. Both were test bugs, fixed in one token each.
Then the campaign: 3 of 8 seeded bugs were caught. The escapes were an unchecked 100% profit-target bound, a 50-trades-per-day cap, a drawdown divisor, and a win-rate scaling in the funded phase that no test exercised. Four new tests close them (one mutant, ev(101) for ev(100), is equivalent: a win probability above 1 behaves like 1). Catch rate on the file went from 42.9% to 100%.