Explainable Graph Markup Language for verifiable AI exchange.
XGML defines a structured reasoning contract: claims, evidence, proofs, and signatures that survive model boundaries and regulatory scrutiny.
Status: planned. English-only demo prompt enforced.
Interoperability is credibility when evidence can be checked anywhere.
XGML is designed for model-portable reasoning artifacts so verification and governance do not depend on a single vendor narrative. Claims, evidence references, and signatures remain inspectable across environments.
Sources: Anthropic RSP v3, Anthropic distillation disclosure, Anthropic zero-days research, and Aevion public evidence matrix.
Portable Schema
XGML payloads move across models and vendors without losing reasoning structure.
Signed Evidence
Proof envelopes support cryptographic signatures for non-repudiable audit records.
Open Governance
Spec evolution is tracked in public with changelogs, proposals, and review workflows.
<xgml version="2.1">
<claim id="c-1042">Veteran is eligible for VR&E intake.</claim>
<evidence ref="m28c.section.iv" confidence="0.97" />
<proof hash="sha256:..." signature="ed25519:..." />
</xgml>Draft
Schema proposal with rationale, examples, and compatibility notes.
Reference
Validator behavior and canonical fixtures formalized in the spec repo.
Interop
Cross-vendor payload exchange validated across gateways and tools.
Ratify
Working group review closes open proposals and publishes release marks.
Validator Track
Run schema checks and signature integrity verification against live payloads.
Open validatorProducer Track
Emit claims, evidence refs, and proofs from your AI application pipeline.
Read producer guideGovernance Track
Participate in change proposals and conformance updates via the public process.
Join governance