Simulation logs for "Behavioral inheritance and evolution in LLM-controlled agent populations"

Raw simulation logs supporting the manuscript "Behavioral inheritance and evolution in LLM-controlled agent populations" (Scott N. Hwang, The Pennsylvania State University). This dataset contains 9 JSON simulation logs (~165 MB total) produced by the Evolutionary Ecosystem, a live agent-based simulation in which populations of LLM-controlled creatures breed, inherit natural-language behavioral instruction corpora ("genomes"), and evolve under cycling environmental conditions. The logs were generated with the BEAR framework (Behavioral Evolution And Retrieval), public release v0.1.8 (https://github.com/snhwang/bear), using the evolutionary_ecosystem example server. Each file is a complete record of one live run, capturing per-creature genomes, births and lineages, expressed behaviors and action markers, epoch and weather transitions, and fitness outcomes. These files are the raw substrate behind the cross-mode inheritance-fidelity comparison and the action-marker decay figure reported in the paper, along with the Mendelian/haploid, diploid co-dominant, and LLM-synthesis breeding conditions. Each of the two large logs contains the full action log, so a single file serves both the inheritance and the marker-decay analyses for its ploidy. (The flee-vs-rally selection logs are smaller and are provided in the code repository below rather than here.) The logs are referenced by the reproducibility package. Evaluation and figure-generation scripts, the smaller committed logs, and the chunked form of the two large logs are available at: https://github.com/snhwang/paper-behavioral-genetics-artifacts (see SIM_LOGS.md, which maps each file to the corresponding figures and analyses). Reproduction runs from the committed logs and does not require this dataset. The two large logs are archived here because their monolithic form exceeds GitHub's per-file size limit. Changes from version 1: Version 1 mistakenly included monolithic logs from an earlier, superseded simulation run whose birth counts do not match the numbers reported in the manuscript. Version 2 replaces them with the logs from the run the paper actually reports (1,278 births for the Mendelian haploid condition, 1,254 for the diploid co-dominant condition). Specifically: - The stale inheritance logs (mendelian_haploid, diploid_codominant) and the separate "v2_action" logs are replaced by two consolidated logs, each holding both the birth log and the action log for its ploidy. - The two second-seed "run1" logs, which are not reported in the paper, are removed. - The seven LLM-synthesis logs are unchanged from version 1. Contents: - sim_log_mendelian_haploid.json (~71 MB, 1,278 births): Mendelian haploid run, includes the action log. - sim_log_diploid_codominant.json (~70 MB, 1,254 births): diploid co-dominant run, includes the action log. - sim_log_llm_synthesis_free_action.json (6 MB, 39 births): LLM synthesis, free action. - sim_log_llm_synthesis_predator_bloom_locked.json (5 MB, 47 births): LLM synthesis, predator pressure. - sim_log_llm_synthesis_abundance_locked.json (4 MB, 43 births): LLM synthesis, abundance scenario. - sim_log_llm_synthesis_free_epoch.json (2 MB, 45 births): LLM synthesis, free epoch, baseline. - sim_log_llm_synthesis_free_epoch_run1.json (2 MB, 52 births): LLM synthesis, free epoch, seed 1. - sim_log_llm_synthesis_free_epoch_run2.json (2 MB, 52 births): LLM synthesis, free epoch, seed 2. - sim_log_llm_synthesis_free_regulated_pop.json (2 MB, 35 births): LLM synthesis, population-regulated. License: Open Core Ventures Source Available License (OCVSAL) version 1.0. Provisional patent pending.