Rotator self-check fixed by separating manual-era drift from real automated regressions

A validation script for a framework rotation engine kept reporting four algorithmic disagreements, even though the primary window-violation count stayed at zero. The author traced the noise to two sources: a seed history that truncated earlier context, and manually chosen framework assignments that did not follow the automated minimum-candidate rule. They confirmed that the automated-era records matched recalculated values and that no typos existed in the seed data. A fix was added so that the disagreement check applies only to automatically maintained dates, and the output went from four disagreements to none with the window-violation count unchanged. The author notes that separating manual and automated states in code is more reliable than manually filtering noise after the fact.