Star-LAFT AE Supplement: Reference VASP Execution Traces for SC26 Artifact Evaluation

Le résumé fourni par la source

Reference execution traces from ten Star-LAFT high-throughput DFT workflowruns, released as the artifact-evaluation supplement for the SC26 paper"Star-LAFT: An Agent-Driven Adaptive Framework for Fault-TolerantHigh-Throughput Materials Informatics". The package exists because VASP and its PAW pseudopotentials are proprietary.It allows a reviewer to inspect the workflow construction, the fault-diagnosisand recovery behaviour, and representative execution results without holding aVASP license. Contents: five agent-enabled runs and five same-structure baseline runs(615 files, about 147 MB extracted). All five agent-enabled runs complete thefull Test_spin -> Coarse_relax -> Relax -> Scf -> Band -> Dos sequence and logbetween one and nine recovery suggestions each; none of the five baseline runscompletes. Each case retains the POSCAR input, the generated workflow script,INCAR and preserved INCAR revisions, KPOINTS, selected VASP text outputs,workflow-status and timing records, and scheduler logs. A top-level README.md,summary.csv, and SHA256SUMS are included for inspection and integrity checking. Exclusions: VASP source code, executables, licenses, and PAW POTCAR files arenot redistributed. Large restart and density files (WAVECAR, CHG, CHGCAR,AECCAR*, PROCAR) and the Band/Dos vasprun.xml files are omitted. Clusterusernames, absolute paths, and compute-node identifiers in retained text fileshave been replaced by neutral placeholders. These are records of past executions rather than a controlled timing benchmark.They document workflow behaviour and do not by themselves establish causalityor reproduce the paper's aggregate statistics. Software: https://github.com/pincher-chen/Star-LAFTArchived at: https://doi.org/10.5281/zenodo.21619301

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.