Margin Play: A Multi-Agent System for Public Policy Analysis in the Brazilian Equatorial Margin
Antonio de Sousa Leitão Filho, Fabrício Saul Lima, Selby Mykael Lima dos Santos, Rejani Bandeira Vieira Sousa, Luís Jorge Mesquita de Jesus, Dennys Correia da Silva, Allan Kardec Duailibe Barros Filho · 2026
Read Margin Play: A Multi-Agent System for Public Policy Analysis in the Brazilian Equatorial Margin by Antonio de Sousa Leitão Filho, Fabrício Saul Lima, Selby Mykael Lima dos Santos, Rejani Bandeira Vieira Sousa, Luís Jorge Mesquita de Jesus, Dennys Correia da Silva, Allan Kardec Duailibe Barros Filho (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02614.txt.
Introduction
Leitão Filho and co-authors simulate oil policy on Brazil's Equatorial Margin with learning agents. The margin is Brazil's next offshore oil frontier. Operations are expected in the Foz do Amazonas basin. Fiscal and territorial links run mainly to Maranhão. That state has the lowest human development index in the federation. The core question is when exploration yields net gains for Maranhão.
Six agents model the conflict. The federal government seeks revenue and energy security. The state seeks welfare under royalty rules. The operator maximizes profit under risk. Regulators ANP and IBAMA hold clashing mandates. Amazonian communities prize land and environment over income. The MARL system trains under a centralized-training scheme. Across 60,000 episodes and six scenarios, the answer turns on the policy regime. One configuration lifts welfare by 17.5 percent with lower environmental liability. Regime choice, not output level, is the lever.
Catalog record
| CREATOR | Antonio de Sousa Leitão Filho, Fabrício Saul Lima, Selby Mykael Lima dos Santos, Rejani Bandeira Vieira Sousa, Luís Jorge Mesquita de Jesus, Dennys Correia da Silva, Allan Kardec Duailibe Barros Filho |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02614
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Margin Play: A Multi-Agent System for Public Policy Analysis in the Brazilian Equatorial Margin Antonio de Sousa Leitão Filho, Fabrício Saul Lima, Selby Mykael Lima dos Santos, Rejani Bandeira Vieira Sousa, Luís Jorge Mesquita de Jesus, Dennys Correia da Silva, Allan Kardec Duailibe Barros Filho Abstract The Brazilian Equatorial Margin (BEM) constitutes the next frontier of offshore oil exploration in Brazil, with operations expected to begin in 2026 in the Foz do Amazonas basin. The BEM assets are fiscally and territorially linked, primarily, to the state of Maranhão, currently the state with the lowest Human Development Index in the Federation (HDI 0.676, IBGE 2022). This gives rise to the central public policy question of this work: to what extent and under what conditions does oil exploration in the BEM generate net positive externalities for the state of Maranhão? The problem is intrinsically multi-agent: the Federal Government seeks tax revenue and energy security; the state seeks regional welfare under constitutional royalty earmarking; the operator maximizes profit subject to operational risk; the ANP and IBAMA represent conflicting institutional mandates; and Amazonian communities assign greater weight to territorial and environmental vectors than to monetary income. We present Margin Play, a Multi-Agent Reinforcement Learning (MARL) system that simulates these tensions under Brazilian empirical calibration (Law 9.478/97, Law 12.858/2013, Petrobras Form 20-F, TCU 2.936/2021, MMA NDC 2024, CPT/INCRA/CIMI 2024) and grounded in classical economic literature (Aschauer–Munnell, Cobb–Douglas, CRRA, Atkinson, van der Ploeg). The system implements six agents under the CTDE paradigm (Centralized Training, Decentralized Execution) trained with BRO-MARL. The results, obtained from 60,000 episodes distributed across six scenarios (three macroeconomic baselines, two policy counterfactuals and one structural transformative regime designated MA-Próspero), indicate that the answer to the central question is conditional on the adopted institutional regime: under the reference baseline, the welfare gain is marginal ( W aval ≈ 1.68 W_{\mathrm{aval}}\!\approx\!1.68 ), whereas the MA-Próspero parametric configuration yields Δ W = + 17.5 % \Delta W={+}17.5\% and Δ R com = + 21.3 % \Delta R_{\mathrm{com}}={+}21.3\% , simultaneously with an environmental liability below the reference ( E amb = 0.048 E_{\mathrm{amb}}=0.048 vs. 0.076 0.076 ). It is concluded that the fundamental problem does not consist in a trade-off between production and welfare, but in the choice of the public policy regime linked to exploration. Keywords Multi-Agent Reinforcement Learning \cdot Brazilian Equatorial Margin \cdot Maranhão \cdot Public Policy \cdot Offshore Oil \cdot Cobb–Douglas \cdot Atkinson \cdot CTDE \cdot BRO \cdot royalties 1 Introduction 1.1 Context and Motivation The Brazilian Equatorial Margin (BEM) is a geological complex extending from Amapá to Rio Grande do Norte, encompassing five offshore sedimentary basins (Foz do Amazonas, Pará-Maranhão, Barreirinhas, Ceará and Potiguar), as illustrated in fig. 1 . Estimates of undiscovered potential from the National Petroleum, Natural Gas and Biofuels Agency (ANP) suggest potential reserves of 10 to 30 billion barrels of oil equivalent [ 1 ] . It is noteworthy that the Barreirinhas basin — whose fiscal-territorial affiliation also falls on the state of Maranhão — presents distinct techno-economic optimization challenges, notably in carbonate matrix stimulation processes, whose viability depends on the integration of reactive flow simulations with Net Present Value (NPV) metrics, directly influencing the cost and risk structures assumed by operators [ 1 ] . Figure 1: Map of the Brazilian Equatorial Margin (BEM) with the five offshore sedimentary basins (Foz do Amazonas, Pará-Maranhão, Barreirinhas, Ceará and Potiguar) and the state of Maranhão highlighted as the primary fiscal-territorial affiliate. The designations employed and the presentation of material in this map do not imply the expression of any opinion on the part of the authors concerning the legal status of any country, territory, city or area. The region became the focus of regulatory tension between 2023 and 2025 with the FZA-M-59 case: Petrobras applied for a drilling licence in the Foz do Amazonas basin; the Brazilian Institute of Environment and Renewable Natural Resources (IBAMA) denied the licence [ 2 ] due to deficiencies in the emergency plan; the Ministry of Environment (MMA), initially aligned with IBAMA, subsequently positioned itself in favour of re-analysis [ 3 ] . The conflict illustrates the complexity of the multi-agent problem: a single project involves fiscal interests (Federal Government), constitutional equity interests (state), operational interests (Petrobras), procedural-environmental interests (IBAMA), technical-regulatory interests (ANP) and socio-territorial interests (Amazonian communities), under legal constraints ranging from the Federal Constitution of 1988 [ 4 ] to specific resolutions such as ANP Resolution 882/2022 [ 5 ] . Following the granting of the drilling licence for the Morpho FZA-M-59 well in October 2025, Petrobras announced an investment of approximately US$ 3 billion for the drilling of 16 wells in the BEM between 2025 and 2029, consolidating the transition from geological promise to operational reality. This development validates the urgency of the analytical question addressed in this study. The central question of this work can be stated as: to what extent and under what conditions does oil exploration in the BEM generate net positive externalities for the state of Maranhão? Maranhão is the state with primary fiscal-territorial affiliation to the Foz do Amazonas and Pará-Maranhão basins, and the one with the lowest HDI in the federation (0.676, IBGE 2022). The resource curse literature [ 6 , 7 ] indicates that the influx of oil revenues produces ambiguous effects: positive via royalties, FUNDEB and GDP; negative via Dutch disease, institutional capture and intensification of territorial conflicts. The question is not purely theoretical: in 2024 the state recorded 363 land conflicts (1st place nationally, CPT [ 8 ] ) and 49 cases of violence against indigenous peoples (CIMI [ 9 ] ). Oil exploration in the BEM will take place within this pre-existing socio-territorial context. Moreover, Hodler, Lechner and Raschky (2023) [ 10 ] , applying a causal forest estimator to data from 3,800 sub-Saharan African districts, empirically confirmed that stronger institutions amplify the positive effect of mining activities on economic development and attenuate their negative effect on conflicts — providing causal machine learning validation for the institutional conditionality that is the central premise of this work. Public policy analysis for this scenario requires analytical tools with four simultaneous properties: mandate heterogeneity, physical-economic coupling, calibration with national data, and robust counterfactual capacity [ 11 , 12 , 13 , 14 , 15 ] . 1.2 Objectives and Contributions Margin Play is proposed as a computational instrument for public policy analysis, and its contributions are organised along five dimensions. The first is a six-agent multi-agent architecture calibrated with Brazilian primary sources, including updated territorial calibration (H-TERR-2) based on CPT/INCRA/CIMI 2024 data ( section 3 ). The second consists of reward functions with heterogeneous functional forms (CRRA, log-Aschauer, Atkinson, Cobb–Douglas, Hubbert) and R_{\mathrm{gov}} with decomposition of Net Current Revenue into five components endogenously derived from Brazilian fiscal legislation. The third is the application of the BRO algorithm [ 16 ] under the CTDE paradigm [ 17 ] , with distributional critic TQC (Truncated Quantile Critics) [ 18 ] . The fourth is the counterfactual analysis across six scenarios ( section 4 ): three macroeconomic baselines, two policy counterfactuals and one structural transformative regime (MA-Próspero), totalling 60,000 episodes. The fifth is the characterisation of the MA-Próspero regime — a parametric configuration of six simultaneous