TQS Framework Selects High-Quality Coding Trajectories for SFT
October 12, 2026
The Trajectory Quality Score (TQS) is a linear-time, LLM-free framework that evaluates coding agent trajectories based on structural convergence rather than final outcome. Using this method to filter 155,144 trajectories allows for more efficient supervised fine-tuning by avoiding the imitation of inefficient patterns.
HOW THIS AFFECTS YOU
●
builderYou can improve agent performance by filtering training data for process quality instead of just task success.
●
researcherThis moves the field toward structural rather than outcome-based evaluation for agentic training.