The authors built PlanAudit-DF, a pipeline that extracts prompt-derived obligations and audits whether a plan, PySpark program, and self-generated tests cover them, producing revision feedback.
Passing self-generated tests do not ensure semantic correctness in DataFrame code generation
Across 30 TPC-DS tasks and three pipelines, test-pass rates reached 30/30 while every generated solution still failed to match its stored reference output.
Top University
Jingjing (May) · Liu · Dishu Yang · Hongbo Qin · Hansong Zhang
Research Digest··2 min read
Liu et al.
§