The Reward Calibrator That Learns the Shape of Its Own Judgment

I hit the wall when one noisy signal kept dragging candidate selection around like a shopping cart with one bad wheel. On paper, the composite score looked stable. Then I watched different weight sets