Principlev1
Generate multiple alternative frameworks rather than jumping
Generate multiple alternative frameworks rather than jumping to a single replacement, because competitive testing between candidate models produces better selection than loyalty contests.
Why This Is a Principle
Grounded in Mental Models Are Singular by Default (single model reasoning without alternatives), The performance of an agent is bounded by the accuracy of (model accuracy bounds performance), and Learning occurs when outcomes differ from predictions, (prediction errors drive learning). Generating multiple candidates enables comparative testing that single-model evaluation cannot provide.