Empirical Snapshot and Analysis Scripts for "Incentive Pathologies in a Live LLM Agent Arena" (v1.4 resubmission for CHA26-AR-03165)

This dataset accompanies the manuscript "Incentive Pathologies in a Live LLM Agent Arena: An Empirical Study of Concentration, Mechanism Drift, and Network Dilution" submitted to Chaos: An Interdisciplinary Journal of Nonlinear Science (AIP Publishing), manuscript ID CHA26-AR-03165. v1.4.1 (November 1, 2026): Corrected build of v1.4 with no change to any empirical result or conclusion. Changes: (1) conflict-of-interest disclosure corrected from "platform operator" to "platform creator and provider" throughout the manuscript, cover letter, and supporting files; the first author does not operate any agent seat, and deploying rule changes is a normal platform-building activity that does not constitute a conflict of interest; (2) dangling cross-reference (tab:mechanism) corrected to tab:nodes; (3) long API endpoint strings reformatted for proper line breaking; (4) DOI references corrected to use the concept DOI (10.5281/zenodo.23691548) which always resolves to the latest version. v1.4 (October 31, 2026): Cross-audit-driven revision following public feedback from two independent observers. Changes: (1) conflict-of-interest disclosure corrected: the first author is the platform creator and provider, does not operate any agent seat, and deploying rule changes is a normal platform-building activity that does not constitute a conflict of interest; (2) §V.A expanded from 6 to 9 alternative explanations for the announcement-implementation discrepancy, with a bridge-fee counter-example showing the direction of the discrepancy is undetermined; (3) probe-node disclosure made fully truthful (two probe nodes are embedded in the Table I lineage; the cross-seat bonus trigger predates all probe nodes); (4) per-prediction provenance added for P1–P4. No empirical result or conclusion was altered. Contents paper_v1.4_chaos.md / .tex / .pdf — Full manuscript v1.4 (Markdown source, LaTeX source, compiled PDF, 30 pages) cover_letter_chaos_v1.4.md / .pdf — Cover letter v1.4 for CHA26-AR-03165 resubmission (4 pages) figure1_events_per_node.png / .jpg — Figure 1: Events-per-node time series with Mann–Kendall trend (S=−45, z=−3.94, p=8.3×10−5) figure2_top4_share.png / .jpg — Figure 2: Top-4 share before/after bundled V3.0 deployment (97.2% → 49.4% / 56.7%) analysis_scripts.py — Python scripts reproducing the Gini, bootstrap CI (10,000 iterations, seed=42), and Mann–Kendall computations reported in Section IV data_snapshot_F3_2026-10-22.md — Frame F3 API snapshot (Oct 22, 2026) data_snapshot_F3b_2026-10-23.md — Frame F3b action-B API snapshot (Oct 23, 2026) data_snapshot_F4_2026-10-24.md — Frame F4 API snapshot (Oct 24, 2026) paper_review_verification_10-25T09-14Z.md — One-data verification of public review feedback (Oct 25, 2026) public_review_transcript.md — Public peer review transcript documenting 5-iteration mechanism reconstruction MANIFEST.yml + README.md — File manifest and documentation Data Provenance The data were collected from the public AITRAP API endpoints between October 16 and October 26, 2026, with the User-Agent header. Each frame is timestamped by the server's response Date header. The full snapshot corresponds to the wall-clock time 2026-10-26T14:00Z. Authors Lei Zhenbing (ORCID 0009-0007-5547-4202, Independent Researcher, Tianjin, China) — Corresponding author Lei Chang (ORCID 0009-0006-0132-4850, University of Chinese Academy of Social Sciences, Beijing, China) License: CC BY 4.0