Keep tests for failure boundaries absent from system runs

Keep focused tests when the maintained system run proves only the successful path and cannot safely force corruption, provenance drift, threshold leakage, malicious filesystem inputs, concurrency cancellation, or metric-denominator mistakes. Those tests protect distinct failure boundaries even when they use small synthetic fixtures. Delete tests that merely restate source-file spelling or constants when an executable build, smoke, or evaluation artifact already proves the same property; otherwise textual snapshots create churn without increasing fault detection. The limit is that synthetic numerical tests cannot substitute for real dependency, accelerator, convergence, or holdout evidence.

Applies when: Auditing a mixed test suite around expensive or operationally gated end-to-end runs. Keywords: test value failure boundaries system runs provenance corruption numerical evaluation