Principlev1
Inventory specific situations where your old schema produced
Inventory specific situations where your old schema produced accurate predictions, then test whether your new schema also handles those cases before replacing the old model.
Why This Is a Principle
Derives from The performance of an agent is bounded by the accuracy of (world model accuracy bounds performance), Learning occurs when outcomes differ from predictions, (learning from prediction error), and Expert performance in complex domains requires deliberate (expert performance requires deliberate practice). This is actionable protocol for TESTING backwards compatibility. Not foundational but derived from how learning and validation work.