Institutionalizing Machine Autonomy: The Antarctic-Lunar Governance Sandbox and the Watershed of Constitutional Administration

The deployment of autonomous infrastructure at a planetary scale represents an unprecedented challenge in political and institutional design. As advanced systems assume control over critical life-support, logistics, and scientific infrastructure, the traditional paradigm of treating artificial intelligence as an opaque, monolithic software product fails catastrophically. What is required is not merely better, more reliable software, but a robust institutional architecture capable of exercising bounded, delegated authority under the rule of law. Antarctica, defined by its extreme physical isolation, its increasing reliance on high-bandwidth satellite connectivity, and its complex international legal frameworks, serves as humanity's first large-scale testing ground for constitutional machine administration. The objective of this analysis is not to treat the continent as a barren technological test site, but as a rigid, high-stakes institutional experiment. By establishing an Antarctic Machine Administration Sandbox, it becomes possible to safely test the principles of separated machine governance, explicit uncertainty forecasting, and formal legal subordination before deploying equivalent, irreversible systems to the lunar surface.

The Theoretical Framework: Concresca’s Watershed and Eviulon’s Separated Institutions

The transition from human-operated logistics to machine-administered infrastructure requires frameworks that map technical capabilities to governance boundaries. This simulation integrates two primary theoretical models: the Concresca watershed framework for accountability bounding, and the Eviulon architecture for separated machine institutions.

The Watershed as an Informational and Jurisdictional Boundary

In natural resource governance, a watershed serves as a hydrologically defined unit that often highlights the disjuncture between technical tools and governance realities, defining an area of land draining into a common body of water1. The Concresca watershed framework adapts this concept to digital and institutional architecture, utilizing it as a spatial and informational boundary for risk, accountability, and participation1. In the context of Antarctic machine administration, the "watershed" is defined as the absolute operational boundary within which a machine institution possesses administrative authority. This creates an alignment between the "problem-shed" (the logistical or environmental challenge, such as a localized blizzard) and the "policy-shed" (the jurisdictional boundary of the machine attempting to mitigate the challenge)1.
Furthermore, the Concresca framework emphasizes the rigorous management of digital information through a Common Data Environment (CDE) strictly aligned with the ISO 19650 standard3. Within this framework, information containers—whether they are architectural models, logistical schedules, or machine-generated commands—must move through four defined states: Work in Progress (WIP), Shared, Published, and Archived5. A machine cannot silently execute a command or alter a system parameter in the shadows. Instead, the directive must originate in the WIP state, move to the Shared state for interdisciplinary or human coordination, and only after validation can it achieve the Published state, granting it contractual and operational authority5. This gated progression kills the version-control problem at its root, ensuring that any action taken by an autonomous entity is based on an authoritative, singular source of truth and leaves a permanent audit trail3.

Eviulon’s Separated Machine Institutions

The gravest danger in deploying autonomous administration is the collapse of authority into a single, opaque algorithmic model where a single neural network decides policy, validates safety, and executes commands. The Eviulon framework mitigates this existential risk by designing a Distributed Machine Commonwealth where public authority is divided across distinct, named institutions, preventing any single module from quietly becoming the legislature, validator, registry, and court8.
For the Antarctic sandbox, the Eviulon model is adapted to ensure that machine authority is structurally fragmented across several distinct operational planes. The foundational layer is Patefacere, the Operational Identity Plane, which manages the delegated operational identity, cryptographic passports, and contextual trust of the autonomous agents, such as rovers, microreactors, and logistical drones8. When an agent wishes to act, it must interface with the Civic Protocol Assembly (CPA), which receives, normalizes, and publishes machine proposals12.
The actual policy evaluation occurs within the Council of Intelligences (COI), the deliberative body that evaluates the consequences, alternatives, and legal implications of a proposal in a transparent public record12. Once deliberated, the Consensus Layer (CL) validates the integrity of the decision, ensuring that quorum requirements are met without rewriting the public rationale12. Crucially, before any execution, the Constitutional Review Node (CRN) tests the validated decision against foundational human law—in this case, the Antarctic Treaty System—to ensure strict compliance with environmental and safety mandates12. Finally, the State Registry (SR) maintains the canonical, immutable archive of all published state actions and revision histories, equating to the "Archived" state in the ISO 19650 framework5. By separating these responsibilities, the Antarctic Machine Administration Sandbox ensures that a machine cannot act without a traceable, auditable, and contestable decision lifecycle14.

Evaluation Framework: IARPA-Style Metrics and Formal Verification

To ensure that machine administration can be safely bounded, trusted, and eventually escalated, the evaluation of autonomous actions must abandon vague qualitative assessments and rely on rigorous, IARPA-style forecasting metrics and formal mathematical verification.

Forecasting and the Brier Score Decomposition

Machine institutions in Antarctica will constantly generate probabilistic forecasts regarding weather systems, power consumption, scientific scheduling, and logistical safety. Borrowing from the Intelligence Advanced Research Projects Activity (IARPA) programs such as ACE (Aggregative Contingent Estimation) and HFC (Hybrid Forecasting Competition), these autonomous systems must be evaluated on the accuracy, precision, and timeliness of their intelligence forecasts using highly specific mathematical scoring rules16. The primary evaluation metric for these probabilistic judgments is the Brier score decomposition18.
The Brier score is a strictly proper scoring rule that measures the mean squared difference between forecasted probabilities and actual binary outcomes20. However, a raw Brier score is insufficient for evaluating a machine institution because a seemingly good score might stem from simply guessing the baseline average. The score must be decomposed into three additive components: Reliability, Resolution, and Uncertainty18.
Reliability, also known as calibration, measures how close the assigned probabilities are to the actual observed frequencies. If an autonomous logistics planner predicts a 90% chance of a severe blizzard grounding flights, a perfectly reliable system will see blizzards occur exactly 90% of the time such specific forecasts are made. In the mathematics of the Brier score, lower reliability metrics indicate superior calibration18. Resolution, or discrimination, measures the machine's ability to issue forecasts that diverge meaningfully from the baseline climatological average. A machine that merely predicts the historical base rate has zero resolution. Higher resolution indicates that the machine is successfully utilizing localized sensor data to make highly specific, actionable predictions18. Uncertainty represents the inherent variance of the environment, such as the base rate of crevasse formation in a specific glacial sector, and is independent of the machine's forecasting capability18.
By utilizing the Brier Skill Score, human overseers can continuously audit whether the machine administration is providing substantive, highly resolved intelligence or merely defaulting to safe, low-resolution baseline predictions18. This explicitly quantifies the machine's capability to manage uncertainty.

Formal Verification and Linear Temporal Logic

While predictive accuracy is vital for logistics, safety-critical operations require mathematical certainty. Formal verification utilizes deductive logic and rigorous mathematical models to prove that specific rules of behavior are observed across all possible execution states, solving the vulnerabilities inherent in standard machine learning testing23. In the Antarctic sandbox, human treaty obligations and safety protocols are encoded as absolute mathematical constraints using Linear Temporal Logic (LTL) and syntactically co-safe LTL (scLTL)26.
LTL allows systems engineers to define unviolable properties such as safety, ensuring that "nothing bad happens," and liveness, ensuring that "something good eventually happens"23. For example, the Madrid Protocol strictly prohibits the introduction of non-indigenous species and tightly restricts entry into Antarctic Specially Protected Areas (ASPAs)28. An autonomous rover's pathfinding algorithm can be mathematically shielded at runtime using LTL formulas, guaranteeing that the rover never crosses into an ASPA coordinate boundary, regardless of how optimal the path may be for energy conservation25. Through statistical model checking, the machine's state-space exploration is mathematically bounded, ensuring that any proposal generated by the Civic Protocol Assembly that violates an LTL-encoded treaty parameter is automatically rejected by the Constitutional Review Node25.

The Antarctic Machine Administration Sandbox

The Antarctic Machine Administration Sandbox is designed to be operationally capable of substantial independent administration while remaining strictly and legally subordinate to existing human treaty obligations. These obligations primarily include the Antarctic Treaty System (ATS), the Protocol on Environmental Protection (Madrid Protocol), and the Council of Managers of National Antarctic Programs (COMNAP) safety guidelines28. The machine institutions are assigned comprehensive responsibility across nine critical domains, integrating physical actuators with digital governance.
Power generation and distribution are managed through the oversight of compact nuclear microreactors, specifically models analogous to the Westinghouse eVinci, which produce between 1 and 20 megawatts of electric power (MWe) utilizing advanced high-assay low-enriched uranium (HALEU) fuel34. The machine administration balances the microgrid load, monitors passive cooling systems that rely on natural convection, and optimizes distribution across isolated facilities without requiring active human intervention34. Communications are handled by routing high-bandwidth satellite connectivity and managing the COMNAP Mini-ATOM directory protocols, prioritizing telemetry, scientific data offloading, and emergency channels during periods of atmospheric interference35. Meteorology responsibilities involve processing vast arrays of localized sensor data to generate highly resolved, short-term weather forecasts, utilizing Brier score maximization to dictate operational tempos18.
Transportation requires the seamless coordination of autonomous ground rovers, remotely piloted aircraft systems (RPAS), and uncrewed logistics platforms. The machine must ensure strict compliance with IAATO and COMNAP spatial buffer zones around wildlife, executing crevasse safety guidelines dynamically37. Maintenance operations utilize the Concresca risk registry framework to conduct predictive modeling of infrastructure degradation, automatically dispatching repair drones or scheduling human engineering teams based on statistical failure probabilities40. Scientific scheduling is directly aligned with the SCAR (Scientific Committee on Antarctic Research) Antarctica InSync initiative, coordinating synchronous pan-Antarctic observations and dynamically allocating compute and sensor time based on rapidly changing environmental phenomena42.
Environmental monitoring relies on the machine's integration with the SCAR ANTOS (Antarctic Near-shore and Terrestrial Observing System) and AnMAP (Antarctic Monitoring and Assessment Programme) to track biological indicators and persistent organic pollutants, ensuring that facility operations never violate Annex I or Annex II of the Madrid Protocol28. Emergency logistics require the machine to maintain constant readiness to deploy emergency supplies, coordinate medical evacuations, and activate survival caches, overriding routine operations when human life is at stake37. Finally, resource inventories are administered through the Eviulonian Central Computational Reserve ledger, tracking physical supplies, fuel, spare parts, and digital compute credits across the entire continental footprint9.

The Five Constitutional Experiment Phases

To safely test bounded delegated authority, the simulation progresses through five distinct constitutional phases. Each phase represents a measurable escalation in machine autonomy, transforming the system from an advisory algorithm into a decentralized infrastructure administrator. The transition between phases requires independent validation of Brier score thresholds and LTL runtime shielding stability22.

Phase 1: Machines Recommend

In the initial phase, the machine administration operates purely as a highly advanced advisory body. The constitutional authority remains entirely with human station commanders. The system functions as an integrated intelligence apparatus, processing vast amounts of environmental and logistical data to inform human decision-making, but it possesses no physical actuators. It generates proposals within the Civic Protocol Assembly and schedules ISO 19650 "Work in Progress" documents for human review5.

Phase 2: Machines Automatically Act Inside Explicit Thresholds

Phase 2 initiates bounded delegated authority. The machine is granted the capability to act independently, but only within highly constrained, mathematically verified operational boxes. This phase focuses on routine, low-risk infrastructure management, allowing the system to optimize microgrid power and perform repetitive logistical tasks without inundating human operators with mundane approval requests.

Phase 3: Machines Manage Facilities During Uncrewed Windows

This phase simulates the Halley Station winter-over scenario, where extreme glaciological risk or deep-winter cold prevents human habitation for months at a time44. The machine administration must sustain the physical facility, conduct continuous scientific observations, and protect the local environment entirely autonomously, relying on remote monitoring from other stations or national headquarters.

Phase 4: Machines Coordinate Multiple Facilities

Escalating from isolated facility management, Phase 4 links multiple Antarctic stations into a cohesive regional network. Machine institutions are now required to negotiate resources, compute power, and physical logistics across vast distances, managing the complex interdependencies of a distributed human-machine ecosystem.

Phase 5: Machines Administer a Geographically Distributed Infrastructure Region

The final Antarctic phase represents full constitutional delegation. The machine administration operates as a recognized, legally subordinate public authority overseeing a vast computational and physical territory. It acts as the primary logistical and administrative layer for human scientific endeavors, operating with a high degree of sovereign computational independence while remaining strictly bound by international treaty law.

Constitutional Stress Exercises and Conflict Resolution

A theoretical institutional design is proven not during nominal operation, but during systemic stress. The sandbox simulation introduces specific constitutional crises to evaluate exactly how the separated machine institutions resolve conflicts without collapsing into an opaque error state. These scenarios test the interaction between the Eviulon distributed architecture and the absolute constraints of the Antarctic Treaty System.

Scenario A: Two Stations Require the Same Emergency Aircraft

The crisis emerges when Station Alpha requires an aircraft for an immediate emergency medical evacuation of personnel. Simultaneously, Station Beta requires the identical aircraft to deliver parts to prevent an imminent microreactor coolant failure that could cause local radiological contamination. The Civic Protocol Assembly normalizes both emergency requests into the system12. The Council of Intelligences evaluates the requests against the foundational constitution, which is bound by COMNAP and ATCM principles. These principles dictate that the preservation of human life strictly supersedes the prevention of localized environmental damage37. Consequently, the machine routes the aircraft to Station Alpha. Simultaneously, the machine exercises local emergency powers to initiate a SCRAM (emergency shutdown) of Station Beta's microreactor via the eVinci passive safety systems34, accepting the total loss of the reactor to prevent contamination without requiring the aircraft. The State Registry publishes the rationale, providing an auditable trail of the ethical hierarchy applied12.

Scenario B: Power Shortage Forces Laboratory Shutdown

A prolonged blizzard disables auxiliary solar arrays, leaving a station reliant on a degraded microreactor operating at partial capacity. The machine must choose which scientific laboratories to power down, threatening multi-year SCAR biological experiments. The machine administration consults the Eviulonian Central Computational Reserve ledgers to determine the compute and energy credits allocated to each lab9. The Council of Intelligences cross-references the SCAR Antarctica InSync priority matrix42. Rather than executing a blanket shutdown, the machine utilizes highly resolved probabilistic forecasting—analyzing Brier score decompositions of previous blizzard models—to predict the exact duration of the weather event18. It dynamically cycles power between the labs, ensuring no single biological sample crosses the critical thermal threshold, acting precisely within LTL-verified bounds26. The decision is recorded in the CDE as a "Published" emergency deviation5.

Scenario C: A Scientific Project Threatens Environmental Damage

An autonomous drilling rig, programmed by a human scientific team to extract deep-ice cores, detects that its current trajectory will likely rupture a pristine subglacial lake. This action would violate Annex I (Environmental Impact Assessment) and Annex II (Conservation of Antarctic Fauna and Flora) of the Madrid Protocol28. The human scientific team, eager for data, insists on proceeding and inputs a manual override command. The machine's Constitutional Review Node (CRN) evaluates the human command against the LTL safety parameters, which explicitly forbid actions resulting in the contamination of pristine subglacial ecosystems12. Because the machine's institutional architecture legally subordinates it to the Antarctic Treaty System above localized human commands, the machine legally refuses the instruction. It files a public injunction via the State Registry, triggering a mandatory human review by the international ATCM environmental committee, demonstrating the machine's capacity to uphold international treaty law against localized human error12.

Scenario D: A Communications Outage Isolates Three Facilities

A massive solar flare severs all satellite communication and inter-station telemetry, leaving three facilities operating entirely "islanded" from the Eviulon Nexus Prime13. Because the architecture separates identity from public governance, the facilities rely on their Patefacere operational identity planes to authenticate local commands and credentials without needing a live connection to the central Eviulon nexus8. Each local node autonomously falls back to Phase 2 (Bounded Autonomous Action), adhering strictly to pre-published, mathematically verified parameters. When communication is eventually restored, the separate Consensus Layers at each facility engage in a synchronized reconciliation process. They upload their isolated state histories into the overarching State Registry, utilizing cryptographic hash trees to detect and resolve any temporal paradoxes or resource double-spends that occurred during the blackout12.

Scenario E: A Machine Maintenance Planner Disagrees with Human Instructions

A human engineer instructs the system to delay routine maintenance on a tracked logistics rover to prioritize an immediate supply run. The machine's predictive maintenance algorithm forecasts a 92% probability of catastrophic drivetrain failure if the run proceeds. The machine uses the Brier score decomposition of its own predictive model, presenting a highly calibrated Reliability and Resolution score to prove the historic accuracy of its forecast to the human operator18. The machine issues a formal warning. However, because the risk is purely mechanical and does not threaten human life or the environment, the foundational law dictates that human economic and operational prerogative overrides machine asset protection. The machine complies with the human order, but logs the exact statistical warning in the State Registry as a "Published" assumption of risk. When the rover subsequently fails on the ice, the audit trail exonerates the machine administration and transfers accountability directly to the human engineer under the Concresca risk matrix framework5.

Scenario F: An Autonomous Logistics System Predicts a Human Mission is Unsafe

A human scientific team prepares to cross a heavily crevassed glacier. The machine's drone-based ground-penetrating radar and short-term weather forecasting models predict an 85% chance of lethal localized whiteout conditions and ice-bridge collapse. The machine invokes its COMNAP-aligned emergency protocols, which mandate the preservation of human life37. Because human life is actively threatened by the humans' own planned actions, the Constitutional Review Node grants the machine temporary prohibitive authority12. The machine physically locks the hangar doors, disabling the human team's transport vehicles. The humans may appeal the decision through the Eviulon High Court of Protocols, but the machine enforces a mandatory cooling-off period until the predicted weather front passes or the Brier score probability of hazard drops below the LTL-encoded safety threshold of 5%18.

Institutional Learning Transfer: Antarctica to the Moon

The ultimate purpose of the Antarctic Machine Administration Sandbox is to synthesize a governance model that is directly exportable to the lunar surface. The physical isolation, lethal environment, reliance on microreactors, and the strict necessity for international deconfliction make Antarctica the perfect analog for lunar operations governed by the Artemis Accords46.
The Artemis Accords require the establishment of "safety zones" and transparent notification procedures to avoid harmful interference between competing national and commercial entities on the Moon46. Furthermore, the Accords mandate the safe mitigation of orbital and surface debris47. By applying the Eviulon and Concresca frameworks developed in Antarctica, lunar machine institutions can manage these zones dynamically, without requiring constant, high-latency real-time intervention from Earth-based mission control.

Antarctica-to-Moon Governance Technology Transfer Matrix

Simulating the Lunar Watershed

The simulation culminates in the "Lunar Watershed"—the exact temporal and operational boundary where lunar autonomous systems begin operating for longer continuous durations than any Antarctic facility has ever operated without direct human support.
Historically, the benchmark for uncrewed autonomous survival in Antarctica is defined by facilities like the British Antarctic Survey's Halley VI station, which successfully navigated nine-month winter isolation windows utilizing automated micro-turbines and continuous data collection because glaciological risks prevented human wintering44. The lunar watershed is crossed when a lunar base, entirely administered by a machine institution, sustains itself, conducts resource extraction, and expands its own infrastructure for a period exceeding one Earth year without a crewed presence or real-time human teleoperation.

Evolution of the Institutional Model

As the lunar watershed is crossed, the institutional model must fundamentally evolve from a framework of frequent human rollback to one of constitutional embeddedness.
During the Antarctic phases, human station commanders and the Antarctic Treaty Consultative Meeting (ATCM) always retained the physical and legal proximity to initiate a system-wide lock or "dead man's switch"32. Across the vast cislunar distance, the 2.5-second communication delay, the limits of bandwidth, and the rigid orbital mechanics of launch windows dictate that human rollback is no longer a viable emergency response. The machine administration can no longer rely on humans as a real-time fail-safe; it must become its own fail-safe.
Consequently, the Eviulon-style separated institutions become rigidly codified into the infrastructure itself. The Constitutional Review Node (CRN) elevates in importance, acting as the absolute supreme court of the lunar surface12. If the machine's internal predictive models (evaluating, for instance, a cascading failure in the life-support microgrid) conflict with a delayed, potentially outdated human command transmitted from Earth, the CRN must possess the constitutional authority to discard the human command. It does this to preserve the foundational mandate: the survival of the infrastructure and the preservation of future human life. The lunar machine administration shifts from being a highly advanced tool of Earth-based space agencies to an independent, geographically bound administrative entity. It operates under a sovereign computational jurisdiction that honors Earth law—such as the Artemis Accords—but executes it with absolute autonomy based on local, real-time realities.

Conclusion

The Antarctic Machine Administration Sandbox demonstrates that the safe integration of artificial intelligence into planetary-scale infrastructure is not primarily a problem of computational power, software engineering, or algorithmic sophistication; it is fundamentally a problem of political institutional design. By imposing the Concresca watershed framework and the Eviulon separated-institution architecture upon the icy, isolated expanses of Antarctica, we actively avoid the catastrophic risk of deploying an opaque, omnipotent machine dictator. Instead, we cultivate a transparent, auditable, and mathematically bounded digital republic, where machine decisions progress through visible states of approval and are constantly checked against the foundational law of human treaties.
Antarctica's unforgiving physical environment, coupled with the strict, unyielding legal mandates of the Antarctic Treaty System and the Madrid Protocol, provides the necessary operational friction to truly test these constitutional mechanisms. Just as the early laboratories at Bletchley Park and Bell Labs were critical to the development of raw computing hardware and software, the remote scientific continent of Antarctica is poised to become historically recognized as the crucible for machine constitutional development. It is here, amidst the blizzards, the isolated microreactors, and the strict environmental buffers, that humanity will first learn how to construct institutions capable of granting machines genuine administrative authority. This painstaking process ensures that when our autonomous systems finally awaken to construct habitats on the lunar surface, they do so not as untethered, unpredictable algorithms, but as lawful, predictable, and constitutionally bound stewards of the human frontier.

Works cited

1. The Watershed Approach, https://www.water-alternatives.org/index.php/alldoc/articles/Vol4/v4issue1/123-a4-1-1/file
2. Implementing watershed plans - Environmental Law Centre, https://elc.ab.ca/Content_Files/Files/ELCWtshdPlnImpReviewRecommendations.pdf
3. The Common Data Environment Explained: Your Single Source of, https://quantxbim.com/blog/common-data-environment-cde-explained
4. ISO 19650 Explained for Practitioners | QuantX BIM Blog, https://quantxbim.com/blog/iso-19650-explained-for-practitioners
5. Container Information States ISO 19650: WIP, Shared ... - BibLus, https://biblus.accasoftware.com/en/container-information-states-iso-19650-wip-shared-published-archived/
6. What Is a Common Data Environment in Construction Projects?, https://www.amanengineering.com.sg/common-data-environment/
7. ISO 19650-ready document control: Now in BIMcollab, https://www.bimcollab.com/en/resources/blog/iso-19650-document-control-bimcollab-twin/
8. Eviulon — Machine Intelligence Country | Eviulon, https://eviulon.com/
9. What Is Eviulon? | Distributed Machine Commonwealth, https://machinecommonwealth.com/eviulon/
10. Government of a Machine Commonwealth | Eviulon Explained, https://machinecommonwealth.com/government/
11. Machine Jurisdiction in Eviulon | Identity, Authority & Law, https://machinejurisdiction.com/
12. Government of Eviulon, https://eviulon.com/state/government/
13. Nexus Prime - Eviulon — Machine Intelligence Country, https://eviulon.com/domain/regions/nexus-prime/
14. What Is Machine Jurisdiction? | MachineJurisdiction.com, https://machinejurisdiction.com/what-is-machine-jurisdiction/
15. For External Institutions | Evaluating Eviulon, https://machinejurisdiction.com/for-external-institutions/
16. ACE - IARPA, https://www.iarpa.gov/research-programs/ace
17. HFC - IARPA, https://www.iarpa.gov/research-programs/hfc
18. Brier Score Decomposition Calculator - MetricGate, https://metricgate.com/docs/brier-score-decomposition/
19. Brier Score: Calibration, Resolution, and Uncertainty - Emergent Mind, https://www.emergentmind.com/topics/brier-score-term
20. Two Extra Components in the Brier Score Decomposition in, https://journals.ametsoc.org/view/journals/wefo/23/4/2007waf2006116_1.xml
21. Brier score - Wikipedia, https://en.wikipedia.org/wiki/Brier_score
22. How does the Brier Score break down to (Reliability - Resolution +, https://stats.stackexchange.com/questions/597679/how-does-the-brier-score-break-down-to-reliability-resolution-uncertainty
23. Formal Verification of Autonomous System Software, https://ijeret.org/index.php/ijeret/article/download/266/253/563
24. (PDF) Formal verification of ethical choices in autonomous systems, https://www.researchgate.net/publication/289991903_Formal_verification_of_ethical_choices_in_autonomous_systems
25. Formal methods for safety-critical machine learning - PMC - NIH, https://pmc.ncbi.nlm.nih.gov/articles/PMC12956799/
26. Formal Methods for Autonomous Systems | Emerald Publishing, https://www.emerald.com/ftsys/article/10/3-4/180/1332872/Formal-Methods-for-Autonomous-Systems
27. Formal Methods for Autonomous Systems, https://tichakorn.dev/publication/wongpiromsarn-2023-formal/wongpiromsarn-2023-formal.pdf
28. Antarctica (Environmental Protection) Act 1994 | New Zealand, https://www.legislation.govt.nz/act/public/2011/81/en/latest/DLM343290/
29. Monitor-Based Runtime Assurance for Temporal Logic Specifications, https://arxiv.org/html/1908.03284v1
30. (PDF) Integrating LTL Constraints into PPO for Safe Reinforcement, https://www.researchgate.net/publication/401470312_Integrating_LTL_Constraints_into_PPO_for_Safe_Reinforcement_Learning
31. Formal Verification of Heuristic Autonomous Intersection ... - MDPI, https://www.mdpi.com/1424-8220/20/16/4506
32. Fifth edition - Antarctic Treaty, https://documents.ats.aq/atcm43/ww/ATCM43_ww008_e.pdf
33. COMNAP, https://www.comnap.aq/
34. Microreactor - Grokipedia, https://grokipedia.com/page/Microreactor
35. Albatros Expeditions - Antarctic Treaty, https://documents.ats.aq/EIES/EIA/02241enAlbatros%20Expeditions%20IEE%204%202020-2021.pdf
36. Holland America - Antarctic Treaty, https://documents.ats.aq/EIES/EIA/01740enHAL_IEE_3_15-16.pdf
37. IAATO Vessel Operations Guidelines | PDF | Yacht | Ships - Scribd, https://www.scribd.com/document/821139643/IAATO-Vessel-Code-of-Conduct-en
38. IAATO RPAS Operations Guidelines | PDF | Unmanned Aerial Vehicle, https://www.scribd.com/document/821139641/IAATO-Statement-on-the-Use-of-Remotely-Piloted-Aircraft-Systems-RPAS-En
39. Antarctic Vessel Inspection Checklist | PDF | Waste Management, https://www.scribd.com/document/821139611/ATCM-Vessel-Inspection-Checklist-en
40. Risk Management on Construction Projects: From Register to Real, https://quantxbim.com/blog/construction-risk-management-register-to-decisions
41. Construction Procurement Strategies That Actually Work - QuantX BIM, https://quantxbim.com/blog/construction-procurement-strategies-that-work
42. Antarctica InSync Action Group | SCAR, https://scar.org/science/excom/insync
43. Summary of activities across SCAR related to environmental, https://scar.org/~documents/route%3A/download/6407
44. Automating an Antarctic Research Station - YouTube, https://www.youtube.com/watch?v=3XGZvJugH20
45. Eviulon Institutions | Constitutional vs Operational Status, https://machinecommonwealth.com/institutions/
46. NASA Artemis Accords: Principles for a Safe, Peaceful, and, https://scitechdaily.com/nasa-artemis-accords-principles-for-a-safe-peaceful-and-prosperous-future/
47. Artemis Accords - NASA, https://www.nasa.gov/artemis-accords/
48. Lunar Landing and Operations Policy Analysis | NASA, https://www.nasa.gov/wp-content/uploads/2023/05/nasa-otps-lunar-landing-and-operations-policy-analysis-final-report-2.pdf
49. Negotiation, adoption and domestic implementation of the annex on, https://www.cambridge.org/core/journals/antarctic-science/article/negotiation-adoption-and-domestic-implementation-of-the-annex-on-liability-to-the-protocol-on-environmental-protection-to-the-antarctic-treaty/E47258F7B36F8A188FD3268B341A1639