benchmarks validated at release go stale as models find valid solutions verifiers reject; the authors call the gap benchmark maintenance debt and ran a continuous-validation workflow (community reports, traces, oracle checks, human-gated repairs) over Terminal-Bench 2.0. The audit found issues in roughly 30% of the benchmark; the repairs, released as Terminal-Bench 2.1, moved accuracy by over 10 points and leaderboard positions by up to three ranks, and fixed wall-clock limits alone closed a 20-point harness gap on one model.