October 13, 2026: the original Qwen 2.5 Coder 14B baseline achieved 84/100 execution-correct answers on sampled Spider questions with zero skipped cases. A later prompt and schema-coverage refinement improved a separate, predeclared ten-question comparison from 6/10 to 7/10: two fixes and one regression, with 9/10 statements executing and zero skips in each arm. The ten questions were disjoint from the original 100 and five development cases. Independent read-only replay confirmed the scores with unchanged grading. The refinement has not been rerun across the original 100; 84/100 remains the historical baseline. These are sampled project evaluations, not an official full Spider score.

Based on source review and saved project artifacts. Recorded measurements apply to the stated model, dataset and configuration; they are not claims of a current production deployment.