Full Reproducibility Archive for "When Semantic Initialization Backfires: A Popularity-Stratified Diagnostic Study of LLM-Derived Item Initialization in Sparse Sequential Recommendation"

This record provides the full Tier-2 reproducibility archive for the manuscript <em>“When Semantic Initialization Backfires: A Popularity-Stratified Diagnostic Study of LLM-Derived Item Initialization in Sparse Sequential Recommendation”</em>, submitted for journal peer review. The archive contains the Tier-1 verification snapshot plus RecBole-format datasets, cached DeepSeek item descriptions, BGE item embeddings, source code, evaluation scripts, cached per-sample NDCG@10 arrays, paired-bootstrap JSON outputs, deterministic re-run artifacts, robustness-check outputs, and figure-generation inputs. The archive supports both fast verification of the manuscript's numerical claims without GPU retraining and full Tier-2 re-execution from the included datasets and cached semantic resources.