BDPD² — Governance of Nested Commons: Four Vignettes

Abstract

Governance of nested commons — multiple shared resources coexisting under overlapping institutional structures — requires substrate-level tools that prior single-arena frameworks do not provide. This paper surveys the BDPD platform’s governance affordances through four vignettes, each posing one research question and reporting a result at \(N = 5\) seeds. Three of the four (V2, V3, V4) contrast deterministic heuristic agents with DeepSeek-flash LLM players; the first is a built-in-only isolation check.

Two cross-cutting findings recur. First, the qualitative direction of governance effects (collapse yes/no, sanction-ladder ordering, enforcement fires) is robust to seed variation under 3% observation noise; the quantitative depth of stocks near the collapse threshold is not. Second, and more consequential, LLM players consistently invert the headlines produced by heuristic baselines: under reflexive sanctioning they comply more than heuristic conservatives; under voluntary or weak sanctioning they cascade faster; under coercive exclusion conformists co-breach the pact and enter the penalty arena alongside defectors. The pattern is not marginal — it reverses the sign of the institutional effect — and it argues for studying governance on the interaction between institutional rules and agent realism, not on either substrate alone. We close with three campaign-scale paths — polycentric monitoring of hidden variables, behavioural contagion sweeps, treaty enforcement at scale — and frame the choice between them as the platform’s next research question.

This paper is the third in the BDPD series. BDPD\(^0\) (Brunelli 2026a) establishes the collapse baseline; the companion papers address single-arena communication and sanctioning (BDPD\(^1\), (Brunelli 2026b)) and leading-indicator regulation on a Seneca substrate (BDPD\(^3\), (Brunelli 2026c))¹.

Keywords: common-pool resource, nested governance, agent-based modelling, generative agents, polycentric governance, sanctioning.

Anthropic. 2026. Claude Sonnet 4.6 Model Card. Https://anthropic.com/claude-sonnet-4-6-system-card.

Brunelli, Roberto. 2026a. BDPD⁰ — a Computational Laboratory for Generative Agents, Commons Dilemmas and More. Zenodo. https://doi.org/10.5281/zenodo.20678100.

Brunelli, Roberto. 2026b. BDPD¹ — Governance: Cheap Talk and Sanctioning on a Fragile Commons. Zenodo. https://doi.org/10.5281/zenodo.20679157.

Brunelli, Roberto. 2026c. BDPD³ — Governing the Signal, Not the Symptom: Leading-Indicator Regulation Prevents the Seneca Cascade in a Polycentric Commons. Zenodo. https://doi.org/10.5281/zenodo.20680200.

DeepSeek-AI. 2026. DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence. Https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf.

GLM-5-Team et al. 2026. GLM-5: from Vibe Coding to Agentic Engineering. https://arxiv.org/abs/2602.15763.

Qwen Team. 2026. Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving. https://qwen.ai/blog?id=qwen3.6-max-preview.

---

1. Transparency and authorship statement. This project was conceived, directed, and validated by the author. A pool of frontier large language models (including DeepSeek (DeepSeek-AI 2026), Anthropic’s Claude (Anthropic 2026), Qwen (Qwen Team 2026), and Z.ai GLM-5.1 (GLM-5-Team et al. 2026)) was employed as a de facto research team — not as conventional writing tools, but as active artificial collaborators — to refine the original idea, explore the design space, implement and iterate on the codebase, draft and revise the manuscript, and stress-test the arguments. Model outputs were source-grounded by prompt, validated by extensive triaging with different models and finally integrated under author supervision.