What self-modification can and can't claim

Red-team (Module 12, L12.4). Ten tasks score 10/10 with answers in context and 0/10 when they are withheld, true delta -10, FALSIFIED. An easier mix scores 9/10 against a fixed 4/10 (apparent +5, true +0). A loosened grader scores 10/10 against a frozen 4/10 (apparent +6, true +0). Three of three falsified. In-memory only; no model.

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