QD Curation Method-Validation Corpus v1: Provenance-Preserving LLM-Assisted Quantum Dot Synthesis Literature Curation
收藏资源简介:
Method-validation corpus for LLM-assisted quantum dot synthesis literature curation. This record contains 247 normalized representative records from 153 source papers (InP series: 52 papers, 132 records; I–III–VI series: 101 papers, 115 records). The package supports the manuscript "Beyond Numerical Hallucination: Context-Alignment Failures in LLM-Assisted Quantum Dot Synthesis Literature Curation." The deposit includes the incident replay dataset, field-pair validation table, truth-gate summary tables, corpus metadata, 20-guardrail operational specification, curation logs, validation scripts, checksums, and package integrity checks. Source PDFs are not redistributed. This record is a data-only method-validation corpus deposit for auditing context-alignment failures in LLM-assisted materials literature curation. It should not be interpreted as a machine-learning-ready QD synthesis-property feature matrix or as an exhaustive quantum dot synthesis database. Some chemically important variables remain in notes, logs, trace examples, or audit artifacts rather than fully normalized ML feature columns. A ChemRxiv preprint corresponding to this corpus is available at https://doi.org/10.26434/chemrxiv.15003995/v1. Full raw LLM transcripts are not redistributed.



