onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction | TickerVault