ResearchBenchmarksSemi-structured Retrieval and Query Optimization on STARK-PRIME (test)Follow41.75MRRTEXTRESNET36.75838.05439.3540.646Feb 9, 2026Evaluation ResultsMethodMethodLinksMRRTEXTRESNETmode=deep residual tuningmode=deep residual tuning2026.0241.75DSPyoptimizer=MIPROoptimizer=MIPRO2026.0241.4TextGradsummarization=falsesummarization=false2026.0241.31TextGradsummarization=truesummarization=true2026.0240.72CoTmode=unoptimized lower...mode=unoptimized lower bound2026.0239.76HBCmode=hierarchical imit...mode=hierarchical imitation learning2026.0236.95