Canonical commitments
Compact UTF-8 JSON uses a closed value set. Transactions, blocks, state, and genesis are committed with domain-separated SHA-256.
Reference implementation · v0.1.0
A deterministic Layer‑1 reference core, reduced to inspectable code.
AIMB-X Protocol is a minimal Python reference for canonical messages, atomic state transitions, resource metering, validator votes, quorum certificates, and block finalization.
Implemented in v0.1.0
Canonical execution and quorum checks.
The code is intentionally narrow. Its job is to make serialization, execution, and quorum rules reproducible—not to present an unfinished network as a live chain.
Atomic execution
Accounts carry nonces and balances. Gas is deterministic, batches are atomic, and invalid input cannot partially mutate the candidate state.
.sync operations
The reference runtime implements transfer, namespaced put, and delete operations with closed schemas and bounded keys and values.
Quorum validation
Proposals are reproduced deterministically. Finalization checks unique vote objects representing at least two-thirds of active validator power; authentication is an integration-layer assumption.
Reference finalization path
Reference finalization sequence.
This is deterministic proposal validation and certificate checking. It is not a live Byzantine-fault-tolerant round or timeout protocol.
Propose
An active validator proposes the next block over the finalized tip.
Re-execute
Validators reproduce the ordered transactions and compare the state commitment.
Certify
Unique vote objects, assumed to be authenticated by the integration layer, must reach weighted two-thirds power.
Finalize
A valid block and quorum certificate become the in-memory finalized tip.
Explicit integration boundaries
What this release does not implement.
These are engineering obligations, not hidden features. In particular, the core assumes that vote objects have already been authenticated by a networking and identity layer.
- Network transport and peer discovery
- Persistent storage and crash recovery
- Validator keys, rotation, and vote signatures
- Live BFT rounds, timeouts, and fork recovery
- Proof verification and proof-system integration
- External-chain clients and bridge infrastructure
AI advisory boundaryDraft RFC 0001
AI may advise. It cannot authorize consensus.
Machine-derived signals may rank admissible transactions, flag anomalies, or recommend timing. They do not determine validity, create validator authority, sign votes, or write committed state. Consensus must remain safe when an advisory system is unavailable, inconsistent, or hostile.
Read Draft RFC 0001 (opens in a new tab)Primary sources
Published specifications.
Each link opens the corresponding document in the public v0.1.0 repository.
System modelReplicated-state-machine components and authority boundaries.Open source (opens in a new tab)ConsensusProposal, validation, weighted quorum, and finalization rules.Open source (opens in a new tab)ExecutionCanonical transaction schema, gas, operations, and state commitments.Open source (opens in a new tab)NetworkingRequirements owned by the transport and identity integration layer.Open source (opens in a new tab)Threat modelByzantine, replay, resource, AI-input, interop, and cryptographic risks.Open source (opens in a new tab)Conformance vectorsPublished v1 inputs and commitments for detecting encoding drift.Open source (opens in a new tab)Security policyPrivate reporting path and supported release scope.Open source (opens in a new tab)
Separate technical proposal
The Market Bridge paper is not this implementation.
The Autonomous Intelligence Market Bridge paper proposes agent registries, capital guardrails, market adapters, settlement, and reputation. Those components are a broader architecture proposal; they are not implemented by the v0.1.0 Layer-1 reference core described on this page.
Read the proposal with that distinction