Principlev1
Write forward-looking predictions from each candidate schema
Write forward-looking predictions from each candidate schema to create testable competition, because prospective prediction testing reveals model quality better than retrospective fit evaluation.
Why This Is a Principle
Derives from Hindsight Bias and Calibration Necessity (hindsight bias makes unrecorded prediction learning impossible), Domain-Specific Calibration Development (calibration from domain-specific feedback), and When estimating future task duration, people naturally adopt (inside view underestimates). Prospective predictions with recorded outcomes enable genuine calibration and model comparison.