My GPS once marked a drive as failed because I turned right four meters before its marker. I reached the right place at the right time. One small deviation, and the system still counted it as a failure. The destination was right. The route differed. What counts as success depends on what the system is measuring. That distinction points to a bigger challenge for AI onchain. Traditional blockchains rely on deterministic execution to keep state transitions consistent. AI outputs can vary between executions, and live web data can change between checks. @GenLayer’s Equivalence Principle lets each Intelligent Contract define what makes a result acceptable. A price feed might require an exact value or permit a defined tolerance. A dispute-resolution contract might accept different wording if the conclusion meets its explicit criteria. The leader proposes a result. Validators assess it under the contract’s validation rules. Different outputs do not automatically qualify as equivalent; acceptance depends on the criteria the contract defines. The goal is not identical answers at any cost. It is consistent agreement on what counts as correct for each task. Should onchain AI require identical outputs, or accept different results when they satisfy explicit contract rules? Pick a side and defend your reasoning.