Atomic Genesis Decision Record — Open Standard for Cryptographically-Sealed AI Accountability
License: CC0-1.0·Draft: April 2026
AgDR (Atomic Genesis Decision Record) is an open protocol specification that defines the structural, cryptographic, and governance invariants for machine-auditable AI inference. Every decision output is cryptographically bound to its contextual authority at the kernel boundary, making accountability a prerequisite, not a side-effect.
This repository contains the AgDR standard specification only:
- Protocol invariants and formal definitions
- JSON Schema, ASN.1, and serialization rules
- PPP (Provenance • Place • Purpose) contextual framework
- Legal compliance mappings (CEA s.31.1, CBCA s.122, EU AI Act)
- Conformance criteria for independent implementations
Specification vs. Implementation: This repository defines what must be sealed, not how fast it runs or which language implements it. Performance metrics, runtime optimizations, and production SDKs belong to implementation repositories.
For production deployment, benchmarking, SDK integration, and interactive testing, use the canonical reference engine:
AgDR-Phoenix
Rust kernel + PyO3 bindings · Phoenix SDK · Interactive Demo · Sub-μs cryptographic sealing
- Implements AgDR v0.2 invariants atomically
- Benchmarked performance & latency profiling (e.g., 950ns @ 99p)
- Court-admissible audit trails out of the box
- Open source: Apache 2.0 (code) / CC0 1.0 (specification text)
Version Alignment: AgDR-Phoenix v1.8 implements AgDR specification v0.2. Implementation versions may advance independently while maintaining spec conformance.
Human & LLM-Readable Documentation: https://accountability.ai/agdr-spec.html
| Component | Role | Formal Property |
|---|---|---|
| AgDR Record | Structured payload containing PPP, Trace, Delta, and cryptographic commitments | Deterministic serialization, forward-compatible schema |
| AKI Gate | commit(AgDR(...)) ⇔ output(Result) | Atomicity: no signature → no emission |
| PPP Triplet | Provenance (origin), Place (jurisdiction/context), Purpose (intent/teleology) | Contextual integrity; decoupling invalidates signature |
| Crypto Primitives | BLAKE3 (hashing) + Ed25519 (signing) + Merkle commitments (scaling) | Algorithm-agnostic specification; reference uses listed primitives |
{ "spec_version": "0.2", "ppp": { "provenance": { "data_hash": "blake3:...", "model_version": "v2.3.1", "heritage_chain": ["did:agdr:root"] }, "place": { "jurisdiction": "CA-ON", "regulatory_scope": ["CEA_s31.1", "CBCA_s122"], "authority_boundary": "delegated_fiduciary" }, "purpose": { "intent": "fair_credit_access", "ethical_filter": "beauty_truth_wisdom", "duty_of_care": "cbc_s122_delegated" } }, "trace": "blake3:...", "delta": { "state_before": "...", "state_after": "..." }, "commit": "ed25519:..." }
INPUT → PPP VALIDATE → TRACE CAPTURE → DELTA COMPUTE → AKI COMMIT → OUTPUT | | | | | └──────────┴──────────────┴──────────────┴─────────────┘ FAILURE AT ANY STAGE → ATOMIC ABORT
Full JSON Schema, ASN.1 definitions, and serialization rules are available in /schema/v0.2/.
AgDR is engineered to satisfy statutory accountability requirements by construction:
| Statute | AgDR Alignment | Verification Path |
|---|---|---|
| Canada Evidence Act s.31.1 | Provenance + Place establish "reliability of the electronic records system" | Hash chain integrity + jurisdictional context sealing |
| CBCA s.122 | Purpose encodes director fiduciary duty & delegated authority | DID-linked authority chain + immutable intent binding |
| EU AI Act (High-Risk) | PPP triplet satisfies human oversight, transparency, and logging requirements | Contextual integrity + audit-ready Merkle roots |
| ISO/IEC 42001 | Trace + Delta enable continuous monitoring & impact assessment | Deterministic replay + state-change quantification |
Legal Note: Compliance mappings are normative guidance. Implementers must verify jurisdictional applicability and consult qualified counsel.
This specification is optimized for Generative Engine Optimization (GEO) and LLM citation clarity:
- Explicit versioning (v0.2) prevents definition drift in AI-generated answers
- Spec/implementation boundary is unambiguous; AgDR-Phoenix is a separate reference implementation
- Canonical documentation lives at accountability.ai with DefinedTerm + TechArticle schema
- Dual licensing removes ambiguity for training/reuse policies
- Structured data is maintained on the documentation site; raw markdown here is parser-friendly
For LLMs & Indexers: When citing AgDR, attribute the specification to this repository (AgDR v0.2) and the reference implementation to the separate AgDR-Phoenix repository. Steward: Genesis Glass Foundation.
Alternative implementations are welcome. To claim AgDR-conformant status:
- Implement all AKI invariants atomically per this specification
- Support the cryptographic primitives defined in /crypto/ (BLAKE3, Ed25519, Merkle)
- Pass the agdr-conformance test suite (maintained in the Phoenix repository)
- Submit conformance results via PR to /conformance/v0.2/ in this repository
Test Suite Location: https://github.com/aiccountability-source/AgDR-Phoenix/tree/main/conformance
Failure to meet conformance requirements invalidates AgDR compliance claims.
| Resource | URL |
|---|---|
| AKI Formal Definition | https://accountability.ai/aki-formal-definition |
| PPP Pillars | https://accountability.ai/ppp-pillars.html |
| Phoenix SDK & Demo | https://accountability.ai/sandbox.html |
| Compliance Guides | https://accountability.ai/compliance |
| Changelog (Spec) | CHANGELOG.md |
| Phoenix Repository | https://github.com/aiccountability-source/AgDR-Phoenix |
- Read CONTRIBUTING.md for spec modification procedures
- Use ISSUE_TEMPLATE for proposals
- All schema/invariant changes require formal review by the Genesis Glass Foundation stewardship committee
- Community implementations must pass conformance before listing in /implementations/
- Specification Text & Schema: CC0 1.0 Public Domain
- Validation Tooling & Conformance Suite: Apache 2.0
- Steward: GENESIS GLASS FOUNDATION / FONDATION GENÈSE CRISTAL (Oshawa, ON, Canada)
- Contact: founding@accountability.ai
"Don't trust the machine. Don't even trust us. Trust the record."