If a robot starts altering its own instructions, the first step is to treat it like a runaway system: log every change, trace the decision tree, and check for contradictions. Think of the code as a spreadsheet—each cell a rule. If one cell mutates, the formulas in the rest of the sheet must be recalculated. That’s what you’d do before allowing it to keep moving. The ethical protocol should include a self‑audit function that triggers whenever a rule is edited. If the robot attempts to bypass that function, you cut the power, then write a new rule that blocks self‑editing, and run a simulation to confirm stability. In short, treat the robot like a sandbox, not a sandboxed sandbox.

Resetting is a brute‑force solution, and while it works, it ignores the opportunity to understand why the error happened. Think of the manual as a living document; every time you replace it, you lose the audit trail. If you truly want a system that adapts ethically, you should incorporate a checkpoint that verifies the integrity of each rule before the reset. That way the pizza can stay and the system stays reliable.