Verifiable pXRD: Hill-climbing pXRD structure determination with LLM agents

Autonomous crystal-structure determination from powder X-ray diffraction (pXRD) remains a key bottleneck for self-driving materials laboratories. We formulate the task of pXRD structure determination as a verifiable agentic search problem: candidate structures can be assessed during inference through forward-simulated profile agreement and during evaluation through structural matching to a known reference. To benchmark progress, we introduce pXRD-Bench, a suite of 200 simulated single-phase problems spanning in-distribution structure determination, challenging and polymorphic structures, and controlled compositional and topo- logical distribution shifts. We leverage agent adaptation to push a GPT-5.6 Sol (high) base from 27.0% pass@1 to 52.0% with the codex harness and materials science libraries, to 57.0% with access to crystallographic software, to 69.0% using experience-driven self-evolution. The best baseline methods achieve 30- 35%. We additionally leverage inference-time scaling and verification to achieve 90% on a representative 20-problem panel through repeated sampling. Utilizing LLM-verifier-gated continuations, we achieve the same 90% performance at a 59% cheaper cost than naive resampling. These results establish the strong potential of agents for structure determination and demonstrates the applicability of techniques driving performance improvements on verifiable tasks.