Architectural Fingerprints Corpus: AI-generated and learner essays across four rhetorical modes in entry-level academic writing

A corpus of 612 academic essays: 384 generated by three leading large language models under matched conditions and 228 written by A2–B1 second-language learners under examination, assembled to map the lexical, keyness, and stance "fingerprints" of LLM-generated academic prose against a developmental learner baseline. This deposit accompanies three related publications, listed under "Associated Publications" below. It contains the AI sub-corpus in full, anonymized learner metadata and per-essay analytical outputs (without raw learner text), the lexicons and dictionaries used, the analysis pipeline code, and the figures and tables reported in the manuscripts.