판 이력 — Evaluating Agentic Learning Harness Capabilities Without Labels via the Scaling Hypothesis | AIChainDay