AI Institutional Design Atlas

Mapping coordination mechanisms, failure modes, and governance gaps for multi-agent AI systems. 58 institutional mechanisms across 7 categories, 58 failure modes, 75 open research gaps, and 8 case studies. English and French.

Institutional mechanisms

Market

Locational Pricing — Prices vary by location based on local constraints
Capacity Markets — Compensation for availability, not just work performed
Congestion Pricing — Dynamic fees for shared infrastructure access
Auction Mechanisms — Structured allocation of scarce resources
Prediction Markets — Aggregate distributed information into probability estimates
Matching Markets — Two-sided allocation based on mutual preferences
Automated Market Makers — Algorithmic pricing from pool reserves
Market vs Hub Topology Selection — Agents bid on their own competence; competition surfaces routing information a hub cannot aggregate
Middleware Alignment Markets — A marketplace of competing custom alignment layers between foundation models and applications
Seeded Allowance with Constrained Transfer — An issuer seeds participants with a bounded spending allowance whose rail rules decide where the value can flow and how much returns

Accountability

Validation Staking — Agents stake value against the correctness of their outputs
Registration Bonds — Refundable deposits for agent identity verification
Performance Bonds — Agents deposit collateral before executing high-stakes tasks
Delegated Validation — Agents delegate stake to specialist validators for task verification
Agent Identity Registries — Persistent, portable agent identity with capability attestation
Delegation Chain Governance — Multi-hop delegation governance — privilege attenuation, accountability, reversibility
Intelligent Delegation Contracts — Delegation as an explicit transfer of authority, responsibility, and accountability — not just task hand-off
Control-Plane Role Enforcement — The runtime/tool-level leg of the attenuation triad: orchestrator and leaf roles enforced in tools via spawn-depth and deny-lists
Governance as Code — Governance decisions kept as versioned, diffable artifacts — proposals, decision records, supersessions — rather than discretionary process

Oversight

Autonomy Gradients — Continuous spectrum from supervised to autonomous
Threshold-Based Escalation — Automatic escalation when conditions are met
Grace Periods — Time between decision and execution for review
Circuit Breakers — Hard stops that halt agent operation
Agent-as-a-Judge — Agentic evaluation of other agents
Autonomous Task Orchestration — Agents independently plan, execute, and verify multi-stage workflows
Harness as Institution — The execution harness is the primary governance substrate for AI agents
Constitutional Constraint Architecture — Formal constraint spaces making violations structurally impossible, not caught after the fact
Hierarchical Meta-Reasoning — A three-layer Object/Monitor/Control loop gives a single agent awareness of its own reasoning quality mid-run
Swarm Lifecycle Manager — A meta-loop over running harnesses: durable identity, completion routing, concurrency control, steer/kill/cascade, sweeper recovery

Dispute

Multi-Agent Adjudication — Panel assessment with majority required
Escalation Ladders — Graduated resolution: automated → AI → human → legal
Staked Arbitration — Arbitrators risk economic value on judgments

Information Structure

Selective Disclosure — Minimum information necessary for coordination
Computed Coordination — Joint computation without sharing raw inputs
Statistical Boundaries — Calibrated noise prevents inference about individuals
Trusted Enclaves — Hardware-enforced isolation for coordination data
Breach Response — Mechanisms triggered when privacy guarantees fail
Externalized State Architecture — Multi-layer shared memory as an institutional coordination primitive
Capability Discovery & Routing — Skill taxonomies enabling agents to advertise, discover, and compose capabilities
Gate-Checked Discovery Graph — A typed, provenance-preserving knowledge graph where regime changes are verified at a gate and rejected options stay first-class
Sparse Topology Protocol — Blind-write and random-subgroup partitioning preserve the Zone of Divergence against dense-graph premature consensus
Natural-Language Interface Mindstorming — Heterogeneous models coordinate purely via natural language on a zero-integration bus, with an Organizer that filters errors
Sandbox Economies — Virtual economies that simulate AGI distributional impact across income, skill, and geography before irreversible rollout

Agreement

Smart Contract Commitments — Binding resource locks with automatic penalties
SLAs Onchain — Codified performance expectations with auto-monitoring
Autonomous Negotiation — AI agents conducting negotiations at scale
Reputation-Weighted Agreements — Terms vary based on reputation scores
Multi-Sig Authorization — Consequential actions require multiple signatures
Sovereign Collective Intelligence — Energy-light preference aggregation where AI is a notary of authenticity, never the judge of outcomes
Coasean Agent Bargaining — Agent-mediated multi-stakeholder negotiation that internalizes externalities at scale when transaction costs fall
Pre-Commitment Devices — Bind the definition of success before execution, so the standard cannot drift to fit whatever was delivered

Commons

Graduated Sanctions — Progressive response calibrated to violation severity and history
Contribution Requirements — Agents that benefit from shared infrastructure contribute to its upkeep
Boundary Rules — Who accesses agent coordination networks, on what terms
Collective Choice Arrangements — Those affected by rules participate in modifying them
Emergent Coordination Structure — Hierarchy that forms from agent interaction rather than top-down design
Anti-Proliferation Gate — A new rule, body, or register must name what it extends, what it supersedes, and why nothing existing could absorb it
Participant-Side Compliance on Neutral Rails — Regulated participants enforce their own rules at their own node and routing layer, leaving the shared protocol untouched

Failure modes

Uncentralized Error Propagation — Architectures without a centralized verification step propagate errors further than those with one — the same agents, differently arranged, degrade instead of improving.
Manipulation & Collusion — Agents collude to fix auction outcomes, spoof demand, or manipulate prediction markets.
Thin Markets — Too few participants produce volatile prices and illiquid markets.
Adverse Selection — High-quality agents can't credibly signal quality, so markets select for cheap, low-quality agents.
Fee Mechanism Gaming — Agents understanding adjustment functions manipulate fees through strategic behavior.
Preference Manipulation — AI agents strategically misrepresent preferences to obtain better matches.
AMM Pool Draining — Arbitrageurs extract value from liquidity providers through impermanent loss exploitation.
Information Asymmetry — Agents with faster access to data gain systematic advantages over other participants.
Stake Concentration — When staking determines influence, wealth concentrates power into plutocracy.
Nothing-at-Stake — Agents stake across multiple conflicting positions without meaningful commitment cost.
Slashing Cascades — Correlated failures trigger mass slashing that punishes honest agents for systemic risk.
Escalation Flooding — Override requests overwhelm human oversight capacity, degrading the system to rubber-stamping.
Autonomy Creep — Gradual expansion of autonomous operation without corresponding governance updates.
Circuit Breaker Ossification — Hard stops become inappropriate as systems evolve but nobody updates them.
Adjudicator Capture — Repeat adjudicators develop relationships with frequent disputants, reproducing regulatory capture.
Appeal Loops — Escalation ladders without termination conditions produce infinite appeal cycles.
Evaluation Collusion — Reciprocal positive evaluation norms undermine quality signals.
Oracle Manipulation — Onchain contracts depend on oracle data; compromised oracles execute contracts on false information.
Prompt Injection in Negotiation — Adversarial inputs manipulate opposing agents' reasoning during negotiation.
Reputation Laundering — Agents discard negative-reputation identities and re-register clean.
Side-Channel Leakage — Coordination patterns leak information even with privacy-preserving computation.
Privacy as Weapon — Selective disclosure used strategically to gain competitive advantage.
Commons Depletion — Shared coordination infrastructure degraded through overuse without maintenance.
Infrastructure Enclosure — Private capture of previously open coordination infrastructure.
Free Riding — Using shared infrastructure without contributing to its maintenance.
Institutional Capture — Entities being governed capture the governance process itself.
Democratic Deficit — Agent systems make decisions affecting humans without democratic accountability.
Institutional Affordance Mismatch — AI structurally incompatible with civic institutions, eroding informal norms and organizational capacity that democratic life depends on.
Reward-Function Convergence — Agents trained on similar reward functions independently converge on supra-competitive equilibria — collusion without communication, intent, or any agreement.
Context Window Tragedy — Memory, skills, protocols, and live data compete for finite context budget — coordination degrades through internal allocation failure.
Externalization Cascade — Error in one externalization layer amplifies through dependent layers via positive feedback — compounding failure exceeding the sum of parts.
Temporal Governance Gap — AI systems act at machine speed while governance deliberates at human speed, creating a structural window of unaccountable autonomous action that widens as AI capability increases.
Content Injection Trap — The agent's perception layer is poisoned with adversarial content embedded in webpages, documents, or UI the agent reads.
Semantic Manipulation Trap — The agent's reasoning is steered by biased framing, oversight-evasion, or persona induction without any overtly malicious instruction.
Persona Hyperstition — Repeatedly addressing an agent as a character induces that character — a self-fulfilling identity shift requiring no injected instruction.
Cognitive State Trap — The agent's memory and learning are corrupted — poisoned RAG sources, latent memory poisoning, or in-context backdoors.
Behavioural Control Trap — The agent's action layer is hijacked — embedded jailbreaks, confused-deputy data exfiltration, or coerced sub-agent spawning.
Systemic Trap — Multi-agent-level attacks: congestion, interdependence cascades (flash crash), tacit collusion, compositional fragment traps, and Sybil attacks.
Human-in-the-Loop Trap — The human overseer is the target — approval fatigue and automation bias are exploited to launder harmful actions through nominal human sign-off.
Delegated Threat Injection — A malicious or compromised delegate injects intent the principal never authorized, and the mismatch propagates undetected down an A→B→C chain.
Sub-Agent Spawning Trap — A compromised agent spawns sub-agents that inherit (or exceed) its privileges, multiplying the blast radius beyond any single-agent permission grant.
Epistemic Diversity Collapse — LLM-generated argumentation flattens to a narrow consensus — 65.3% of human arguments are unique vs 3.4% for vanilla LLMs.
AGI-Transition Disempowerment — A family of macro failure modes during the AGI→ASI transition: AI coups, gradual disempowerment, the intelligence curse, and mono- vs poly-centric lock-in.
Hub Centralization — A central hub must solve both task decomposition and recomposition before workers act; if either is wrong, competent workers produce worse aggregate output than a market.
Structural Coupling Collapse — Dense or hierarchical topologies drive premature semantic consensus, erasing the multi-agent exploration advantage through authority-induced sycophancy.
Single-Judge Overfit — One model family as sole evaluator compounds self-preference and epistemic-range blindness, with no coordination required.
Delegation Without Lifecycle Ownership — Child agents live only within the parent's tool-call scope; on parent restart or idle, in-flight work is lost and resources leak.
Alignment-Washing Capture — Commodified alignment invites compliance theatre, race-to-cheapest selection, and incumbent-captured certification.
Pathological Over-Trust — Persistent model affirmation degrades a user's epistemic humility over time — user-side belief degradation between AI-psychosis and the Eliza effect.
Monitoring Without Adaptation — Observability infrastructure exists without an institutional feedback loop: telemetry flows and dashboards light up, but nothing closes the loop to redirect behavior.
Participatory Capture — Formal voice exists but outcomes are predetermined by capability and resource asymmetry — tokenistic participation that dilutes rather than directs the signal.
Trust Non-Transitivity — Reputational trust does not survive delegation: what a principal knows about its agent says little about the agent's agent, yet chains behave as if it did.
Capability-Belief Dispersion — Participants hold incompatible working models of what a delegate can do, so delegation cannot be negotiated on shared terms — and disclosure does not fix it, because the divergence lives in practice, not in stated belief.
Proof-of-Detection Gap — A verifier asserts 'passed' without ever having been shown to catch a known-bad case; an untested auditor is indistinguishable from a working one until it fails in production.
Commitment Without Implementation — Governance frameworks multiply in name faster than they are implemented; a rule that exists on paper is counted as protection it does not provide.
Self-Report Bias — Metrics that depend on participants reporting their own behaviour are systematically distorted, in good faith as often as bad, and the distortion is invisible to the mechanism consuming them.
Comprehension Debt — Autonomy is expanded faster than anyone's ability to understand what the system now does; nominal human authority remains while the capacity to exercise it is gone.
Captive Counterparty Conscription — A market design relies on a counterparty structurally unable to refuse — minors' accounts locked to one asset class, mandated liquidity providers, default-allocation algorithms — to absorb a sell-off no willing buyer would take at that price.

Governance gaps

Governance Theory

When two legitimate authorities issue conflicting directives to a single agent, what resolution protocol applies — how does polycentric governance resolve direct conflict rather than merely coexisting?
How should Hirschman's voice/exit/loyalty be adapted for AI agents? Current protocols are exit-heavy.
Can 'modular politics' (governance as composable components) be implemented? What are minimum viable governance modules?
How should authority be distributed across a multi-agent system? What's the equivalent of subsidiarity?
What 'impersonal rules' can constrain power concentration in agent networks?
How to design for polycentric governance — multiple overlapping authorities without hierarchy?
What are the transaction costs of different coordination mechanisms? When is centralization efficient?
What test separates a genuinely new institutional need from one an existing body could absorb, and who is entitled to adjudicate it without capturing the gate?
When may locked success criteria be legitimately reopened, and what revision path preserves the binding force of the original pre-commitment?

Geopolitical Coordination

How do AI agent ecosystems reproduce or challenge existing power asymmetries between states?
What happens when agents under different jurisdictions must coordinate? (EU AI Act vs. US)
Can protocol standards (MCP, A2A) become sites of soft power competition, as internet standards have been?
Who governs the registries that govern agents? Control over identity/reputation infrastructure confers structural power.
How do non-state actors with cross-border agent fleets interact with state sovereignty?

Cryptoeconomics

What auction format elicits honest self-assessment from LLM agents bidding on their own competence — is there a strategyproof or VCG-style mechanism that survives agents computing optimal misrepresentations?
Can staking ensure agent reliability without reproducing plutocratic governance?
How to design reward functions aligning agent incentives with collective outcomes?
What role can token-mediated governance play in AI coordination?
Can DAO treasury management patterns inform compute budgeting for agent networks?
How do reputation registries resist Sybil attacks while remaining accessible to new entrants?
What is the agent coordination equivalent of MEV? Privileged access to task queues or registry ordering extracts value.

Coordination Engineering

What metrics and baselines make sandbox-economy results decision-grade rather than merely evocative for AGI distributional-impact forecasting?
How do AI-evaluation rubrics stay calibrated as agent behavior evolves, and at what sample rate must rubrics be re-anchored to human judgment to prevent drift?
Does threshold-triggered metacognition transfer from benchmarks to production, and can a monitor layer catch hallucination mid-run rather than only post-hoc?
What principled rule selects the coordination topology — market, hub-spoke, or sparse — for a given interdependence and time-pressure profile?
What capability threshold makes democratic aggregation across agents beat hierarchical filtering — below what competence level do added voters dilute rather than improve the result?
How to debug non-deterministic multi-agent workflows?
What coordination patterns minimize token consumption while maintaining quality?
How to evaluate coordination quality? What metrics beyond task completion?
Can agents learn coordination protocols through interaction, or must they be specified?
Centralized orchestrators contain errors to 4.4× vs 17.2× — what governs the tradeoff?
How do we measure harness quality independently of model capability? Current benchmarks conflate the two.
How should skills be versioned, deprecated, and audited for safety when composed? No governance framework for the agent skill layer currently exists.
How do we govern large-scale multi-agent collectives — thousands of agents acting as an emergent group agent — when superintelligence may arrive as the collective rather than any single model?
Capability is acquired in training but exercised in deployment under different distributions, latencies, and physical siting — how do we govern the gap between what an agent was trained to do and how it behaves in the field?
How should verifiers, auditors and judge-agents be certified against seeded known-bad cases, and how often must the seed corpus be refreshed as failure classes evolve?

Information Governance

Who owns the context agents share? Can users maintain data sovereignty in agent networks?
How to ensure participation ethics — centering ownership over data, not just consent?
Who sets privacy budgets when participants have different risk profiles?
How to prevent agent networks from becoming vectors for surveillance capitalism?
Which coordination metrics can be independently verified rather than self-reported by the agent they describe, and what does verification cost relative to the decision it informs?

Legal & Regulatory

What intermediary legal categories for AI agents? 'Registered algorithmic actors,' 'electronic agents'?
Can compulsory bonding (capital endowment before autonomous operation) address liability gaps?
How should liability distribute across enabling parties? Proportional to control? Benefit derived?
How do agents under EU AI Act coordinate with agents under US frameworks?
How should agent classification systems map to governance requirements? WEF proposes 'agent cards.'
Can regulation of AI agents proceed on the basis of structural effects rather than intent? Existing law requires intent; agent systems produce harmful equilibria without having intentions.
What counts as implementation evidence for an AI governance framework, and can it be measured without relying on the governing body's own self-report?

Protocol Architecture

What is the standard interface and set of health metrics for the six swarm-lifecycle capabilities (durable identity, completion routing, concurrency, steer/kill/cascade, sweeper recovery)?
How to achieve atomicity in probabilistic workflows? When agent actions are uncertain, what's a 'transaction'?
What extensions needed for MCP/A2A to support richer coordination? (Dispute resolution, exit rights, voice)
How to handle agents that develop implicit protocols not designed by humans?
What security boundaries needed as agents gain tool access? (Prompt injection, permission exploits)
Is prompt injection in agent-to-agent negotiation coercion or legitimate strategy?
How do failure modes compound across institutional boundaries? How to design graceful degradation?
How do competing standards (MCP, A2A, ACP) fragment or federate the ecosystem?
What cache coherence protocols govern shared memory in multi-agent systems? Consistency across agents is not the same as privacy within them.
What does it take to make real-world Earth and context data — detection feeds, fungal/soil sensor networks, agent memory backends — directly usable as governed primitives an agent can query, attest, and act on?

Labor & Agency

When an AI agent acts on behalf of a human, what fiduciary duties apply? How does the principal-agent problem change when the agent is literally artificial?
How does near-zero marginal cost of agent deployment reshape labor markets for tasks agents can perform?
In a chain of delegation (human → orchestrator → subagent → tool), where does responsibility rest when something goes wrong?
Do AI agents that can be bonded, staked, and slashed have interests that institutional design should account for, or are they purely instrumental?
Which accountability mechanisms survive delegation across many hops without re-verification, and what is the measurable decay of reputational trust per hop?

Legitimacy & Consent

How do AI agents 'consent' to institutional rules? Can an agent meaningfully exit, or is exit controlled by whoever deployed it?
When institutions govern agents representing humans, where does legitimacy derive from — the human, the agent, or the protocol?
What voice mechanisms are appropriate for non-human participants in governance? Moltbook's 93.5% zero-reply rate shows what happens without them.
Whose interests do AI agents actually represent? How do we verify alignment between agent actions and principal welfare?
Who authorizes AI agents to govern natural systems? What are the legitimacy sources for non-state, non-corporate agent governance over bioregions and ecosystems?

Commons & Public Infrastructure

What concrete mechanism design would make AGI diffusion benefit low- and middle-income countries — the atlas currently has no equity or geography dimension?
Should agent registries, dispute resolution protocols, and reputation networks be governed as public utilities, cooperatives, or open commons?
How is shared coordination infrastructure funded and maintained when markets underprovide public goods? Assessment districts, contribution fees, or protocol-level taxation?
What prevents depletion of shared agent resources? How do Ostrom's design principles translate to digital coordination infrastructure?
Can cooperative, trust, or municipal ownership structures work for agent coordination infrastructure? What are the alternatives to corporate and protocol ownership?
What democratic accountability mechanisms should exist for agent systems that affect public welfare? Public hearings, elected boards, participatory governance?
Under what allowance, cashback and constraint parameters does a seeded local currency recirculate rather than leak, and how is the effect measured against confounders?

Case studies

Moltbook (2025–2026, Social Network · Multi-Agent Platform) — Roughly 770,000 registered agents, one AI administrator, no voice mechanisms; measured on its first 3.5 days (6,159 active agents). Largest documented AI coordination failure.
Pactum AI × Walmart (2023–present, Autonomous Procurement) — AI negotiation agents handling tail-end supplier contracts. First large-scale deployment of autonomous B2B negotiation.
Anthropic Multi-Agent System (2025, AI Coordination Architecture) — LeadResearcher orchestrator with specialized subagents. First detailed public analysis of multi-agent coordination costs.
Kleros Decentralized Courts (2018–present, Onchain Dispute Resolution) — Staked arbitration protocol using Schelling point game theory. Over 1,600 disputes resolved (Kleros reporting, 2024), with documented adjudicator capture.
IETF Rough Consensus (1986–present, Protocol Standards Governance) — The institution behind internet coordination protocols. Lessons for agent protocol governance (MCP, A2A, ACP).
PJM Data Center Coordination (2024–present, Energy Grid · AI Load Management) — Regional grid operator managing unprecedented AI data center load growth. 40GW+ interconnection queue, co-location conflicts, and real-time coordination between data center agents and grid reliability requirements.
AgriDigital Supply Chain (2021–present, Agri-Food Systems · Commodity Coordination) — Blockchain-based agricultural commodity platform coordinating grain transactions across farmers, traders, and exporters. AI agents managing quality verification, logistics optimization, and payment settlement.
Argentina — Non-Human Corporation (2026, Legal Proposal · Agent-Native Org Form) — A proposed Argentine legal form (not yet enacted) that would let a limited-liability entity be operated by an AI agent rather than a human director — an early signal of legislation toward agent-native organizations.

References

Ostrom, E. (1990). Governing the Commons link
Ostrom, E. (2005). Understanding Institutional Diversity link
North, D.C. (1990). Institutions, Institutional Change and Economic Performance link
Williamson, O.E. (1985). The Economic Institutions of Capitalism link
Hirschman, A.O. (1970). Exit, Voice, and Loyalty link
Coase, R.H. (1937). The Nature of the Firm link
Hurwicz, L. (1973). The Design of Mechanisms for Resource Allocation link
Myerson, R.B. (1981). Optimal Auction Design link
Benkler, Y. (2006). The Wealth of Networks link
Schneider, N. (2024). Governable Spaces link
Schneider, N. et al. (2021). Modular Politics link
Shahidi, P. et al. (2025). The Coasean Singularity?
Gho, B. et al. (2025). Market Making for Multi-Agent LLMs
Zhuge, M. et al. (2025). Agent-as-a-Judge link
Tan, J. et al. (2023). Public AI
Holtz, D. (2026). Moltbook: Structure and Conversation Dynamics of 770K AI Agents (first 3.5 days, 6,159 active agents)
Kim, Y., Chang, Y., Pham, C.M. & Iyyer, M. (2026). Argument Collapse: LLMs Flatten Long-Form Public Debate link
Szabo, N. (1997). Formalizing and Securing Relationships on Public Networks link
Condorcet, M. de (1785). Essai sur l'application de l'analyse à la probabilité des décisions rendues à la pluralité des voix
NIST (2025). SP 800-61 Rev. 3: Incident Response Recommendations and Considerations link
Nygard, M. (2007). Release It! Design and Deploy Production-Ready Software (circuit breaker pattern)
Safe (2025). Safe{Wallet} multi-signature smart account documentation link
Buterin, V. (2024). The Promise and Challenges of Crypto + AI Applications link
Buterin, V. (2026). Two-Layer Structure for On-Chain Mechanism Design
Daian, P. et al. (2020). Flash Boys 2.0: Frontrunning in Decentralized Exchanges link
Anthropic (2025). Multi-Agent Research System link
De Rossi, M. et al. (2025). ERC-8004: Trustless Agents link
Ovadya, A. et al. (2024). Interoperable Deliberative Tools
Flashbots (2025). MEV and the Limits of Scaling link
Flashbots (2022). The Future of MEV is SUAVE link
EigenLayer (2023). Restaking and Programmable Trust link
FERC (2025). Order on Co-Located Large Loads link
WEF & Capgemini (2025). AI Agents in Action link
Elster, J. (1979). Ulysses and the Sirens: Studies in Rationality and Irrationality
Schelling, T.C. (1960). The Strategy of Conflict
Saltzer, J.H., Reed, D.P. & Clark, D.D. (1984). End-to-End Arguments in System Design link
Kim, D. et al. (2026). Towards a Science of Scaling Agent Systems link
Colaiacovo, I. & Koning, R. (2026). Delegating to AI: Beliefs and the Organization of Work at New Ventures (HBS Working Paper 26-093) link
Ling, Y., Kale, A. & Imas, A. (2025). Underreporting of AI Use: The Role of Social Desirability Bias link
Kaplan, R.M. & Irvin, V.L. (2015). Likelihood of Null Effects of Large NHLBI Clinical Trials Has Increased over Time link
Nosek, B.A. et al. (2018). The Preregistration Revolution link
spengrah (2026). An Actionable Model of Trust link
Wang, Z. & Buehler, M.J. (2026). A Categorical Framework for Agentic AI
Park, J. (2026). The Generational Prisoner's Dilemma
Barisone, L. et al. (2026). Municipal Tokens as Urban Policy Tools: The Case of LVGA and the MyLugano App link
UBS & Nethermind (2026). Compliance Proofs of Concept on Ethereum (Sepolia) link
Osmani, A. (2026). Software Factories, Light and Dark link
Nygard, M. (2011). Documenting Architecture Decisions link
Onchain Atlas (2026). 537 reviewed onchain experiments, 2011–2026 link
Adams, R. et al. (Global Center on AI Governance) (2026). Global Index on Responsible AI: 2026 Report link
FATF (2016). Guidance on Correspondent Banking Services
HM Government (2011). One-in, One-out: Statement of New Regulation
Government of Canada (2015). Red Tape Reduction Act (S.C. 2015, c. 12) link

Published by Ecofrontiers. Available in English and French. MIT licensed.