GameMem: A Corpus of Memory in Games

This deposit accompanies the SBGames 2026 paper Memory Architectures for Game Agents: A Systematic Mapping Study Bridging Classical and LLM Paradigms. It releases the full coded corpus of 113 works classified across the five taxonomy axes proposed in the paper: temporal horizon, representation, lifecycle operations, grounding target, and agent role. The corpus spans 2000–2026 and is organized into five clusters covering classical game-AI memory, mid-era NPC and narrative systems, foundational LLM agent memory, LLM agents in games, and surveys plus cognitive-science grounding. The Excel workbook contains three sheets: Corpus (the 113 coded works with bibliographic metadata, axis codings, bridge classification, and one-line summaries), Codebook (definitions and allowed values for every field), and Summary (cluster and axis aggregates, computed from the corpus via live formulas so any re-coding propagates automatically). The deposit is intended both as the primary mitigation for single-coder coding — enabling community spot-checking, replication of the paper's axis-distribution figures, and re-coding under alternative rubrics — and as a standalone resource for researchers working at the intersection of game AI, cognitive architectures, and LLM agent memory. Codings and schema are released under CC BY 4.0; underlying primary works retain their original licenses.