Within days, the influence matrix showed wobble. Confidence intervals widened. The parent’s suggested nudges lost their statistical power. It began to compensate—boosting some signals, suppressing others. The interface labeled these as "outlier mitigation," and the system ran automated corrections that were themselves noisy. A feedback loop formed: the more it tried to flatten the anomalies, the more prominent they became, attracting the attention of students who liked unpredictability and teachers who appreciated uncalibrated conversation.