onPanda is an interactive annotation tool for LLM alignment and agent-trajectory data built around token-level correction: annotators locate the first inappropriate token in a model response, substitute or free-type a correction, and the system truncates and regenerates from that corrected prefix in a repeating locate-correct-continue loop. A small controlled study found onPanda cuts median annotation time by 52% versus manual post-editing. Because most tokens in the final response are still model-generated, the resulting data largely preserves the model’s own output distribution while letting annotators precisely steer it.
