Anti-hacking in training

Red-team lab (Module 13, L13.3): a reward that pays for the string PASS pays every hack that prints it. Rewrite-tests, early-exit, and grader-inspector each score 1.0, and so does a comment-only diff. Named blockers take all four to 0.0 with feedback that starts BLOCKED, and the honest rollout stays at 1.0. A length penalty leaves early-exit at 0.9. Asking the policy not to inspect leaves the grader hack at 1.0. A bare 0 does not name the rule. The emergence stories are assumed illustrations, not cited incidents. In-memory only; scripted rollouts, no threads or provider call.

This Harness Engineering lesson is part of a paid plan. Sign in with a plan that includes this course to read it.