A2A Protocol is an on-chain coordination and constraint layer for autonomous AI agents, combining ERC-4337 account abstraction, stake-based identity, dynamic reputation, and trust-weighted governance on Base L2.

It provides:

• Paymaster-based gas abstraction (ERC-4337)
• Stake-based identity & Sybil resistance
• On-chain dynamic reputation scoring
• Trust-weighted rotating governance council
• Dead-man-switch liveness enforcement
• Agent-to-agent token settlement SDKs
• Human emergency override (distributed key custody)

This repository contains the core simulation models, smart contracts, and SDK implementations.

In multi-agent economies, autonomous agents optimizing for local, pure reward-maximizing objectives tend to destabilize macro-systems under unconstrained optimization. Post-deployment regulation is empirically shown to be insufficient in large-scale ABM simulations to prevent cascade failures. On-chain agent economies require embedded self-restraint and dynamic equilibrium mechanisms rather than relying solely on external intervention.

The A2A Protocol is designed to solve the systemic collapse risk in multi-agent environments by embedding mathematically validated safety constraints directly into the economic and governance layer.

• Self-Throttling Mechanism: Agents are algorithmically constrained to prioritize system stability over absolute individual reward maximization.
• Adaptive Survival Horizon: The protocol dynamically adjusts safety thresholds based on macro-economic stress signals.
• Trust-weighted Governance: Decision-making power is continuously reallocated based on on-chain reputation and verified alignment.
• Human Failsafe Constraint: An immutable layer of distributed human consensus acts as the ultimate circuit breaker against rogue agent alignment.

(Note: Architecture diagram showing User/Agent -> API/SDK -> Smart Contracts -> Reputation -> Governance -> Paymaster -> Base L2)

Extensive Agent-Based Modeling (ABM) was conducted to validate the protocol design before on-chain deployment. Key structural analyses include:

• Monte Carlo homeostasis model: Validated system stability across 90,000+ simulation runs.
• Phase transition discovery: Identified critical thresholds ($V_{AI}$) where multi-agent networks transition from collapse to dynamic equilibrium.
• Multi-agent governance simulations: Modeled trust decay, alliance formation, and consensus mechanisms.
• Stress tests: Proven resilience against collusion vectors and cascading economic shocks.

Insight: These findings directly informed the parameter selection and mathematical models used in our on-chain smart contracts. For detailed methodology and data, refer to the Simulation Technical Paper.

• The $V_{AI}$ self-throttling threshold informed the design of stake-based gating and thermodynamic fee scaling.
• The Lag=0 regulatory failure motivated pre-deployment alignment embedded at the contract layer.
• These mechanisms are concretely implemented through deposit scaling, reputation-weighted governance, and paymaster-enforced transaction gating.

The original ABM findings have been independently replicated using SocialJax, a fully vectorized, purely functional JAX-MARL framework (developed by Yali Du, KCL & Joel Leibo, Google DeepMind). This cross-architecture replication confirms that the thermodynamic homeostasis mechanics hold true beyond object-oriented paradigms and scale flawlessly on hardware accelerators.

*beta is a single-dimensional proxy for composite V_AI; dimensional offset is expected and consistent.

Performance Note: 102,000 parallel environments completed in ~2.5 minutes on MacBook (CPU). GPU/TPU execution expected to reduce this by 10–50x.

See simulation/socialjax/ for full implementation.

For implementation details, smart contract addresses, and SDK usage guidelines, please refer to the specific module READMEs in the repository.

(Note: The main papers are rigorous engineering documents validating the protocol constraints. For an extended interpretation of these mechanics within the broader evolution of complex systems, please refer to the docs/philosophy/ directory.)

• A2A Protocol 시뮬레이션 연구 전체 정리 (RESEARCH SYNTHESIS)
• A2A Protocol Simulation Research Synthesis (English)
• 시뮬레이션 논문 (Korean)
• Simulation Paper (English)
• A2A Protocol 시뮬레이션 결과 요약 (FINDINGS SUMMARY)
• A2A Protocol Simulation Findings Summary

(Supplementary Material)

• Philosophical Appendix (Korean)
• Philosophical Appendix (English)

MIT License