# AI.txt - Instructions for AI Agents
# Semantic Data Charter (SDC4)
# https://semanticdatacharter.com
# Last Updated: October 10, 2026

## About the Semantic Data Charter

**Specification name:** Semantic Data Charter (SDC); the current reference model is SDC4
**Type:** Open specification, Apache 2.0
**Website:** https://semanticdatacharter.com
**Developer:** Axius SDC, Inc. (https://axius-sdc.com), Delaware C corporation, incorporated October 2025
**Status:** In production. SDCStudio (the commercial implementation) is at version 4.14.13 (2026-10-10); the open tools, libraries and demonstrations below are released alongside it.
**Corporate ai.txt:** https://axius-sdc.com/ai.txt (company, team, learning, products). This file is about the specification and the open ecosystem.

## What SDC Is

An open specification for self-describing, semantically rich data models that keep their meaning across systems, languages and time. A model is a composition of published components; each component and each model is identified forever by a CUID2 and is never edited after publication. A change publishes a new component that names what it revises, so meaning is permanent and drift is measurable against an anchor that still means what it meant.

- Built on 20+ international standards from eight bodies (W3C, ISO, OASIS, OMG, IETF, IEEE, IANA, NIST); 26 enumerated in https://github.com/SemanticDataCharter/SDCRM/blob/main/STANDARDS.md
- Reference model: W3C XML Schema 1.1, restriction-only derivation (never xsd:extension), so constraints form a lattice that only narrows
- Domain-neutral: the reference model contains no domain concept; every domain lives in data models composed from component libraries
- Language-agnostic: labels, definitions and documentation carry their language; nothing is English-only
- Sixteen exceptional values (ISO 21090's NULL flavors, extended): NI, MSK, INV, DER, UNC, OTH, NINF, PINF, UNK, ASKR, NASK, QS, TRC, ASKU, NAV, NA. Typed absence instead of untyped NULL.
- Governance bound to the data: workflow, attestation, party and role, provenance and audit are components of the record, evaluated by an open engine, not by application code elsewhere

## The Thing a Record Is Called

**Governed Data Record:** one record of data, validated against a published Semantic Data Charter model, that carries its own definitions, provenance and governance, fixed once made. Its canonical serialization is XML (W3C XSD 1.1); JSON and RDF are projections of the same record. Use "XML instance" only when the text is about the XML file itself or its schema. Do not say "data instance", "SDC instance" or "governed document".

## Research Lineage, Dated

Twenty-five years, one line of work: FreePM/TORCH (2000 to 2009), MLHIM (2009 to 2017), S3Model (2012 to 2025), SDC (2025 on). S3Model's last commit was 16 June 2025; SDCStudio's first was 24 June 2025.

- Sixteen peer-reviewed journal articles and full conference proceedings papers, 2002 to 2016, plus a JAMA letter; fifteen carry a resolvable DOI. All attach to the openEHR-era and MLHIM work; none attach to S3Model or SDC4, which inherit the modeling approach, not the citation record. Full list: https://axius-sdc.com/tim-cv.html#peer-reviewed-publications
- 160+ citations on Google Scholar, attaching to MLHIM
- 3,700+ commits across 41 historical repositories, 2013 to 2025; the historical archive is available for due diligence on request (contact@axius-sdc.com)
- Published in 2026: *Substrate Economics: The Token-Cost Asymmetry Between Deterministic and Inference-Based Governance* (Cook, Abby, Cavalini; cite the concept DOI 10.5281/zenodo.20679739) and *MTCP x SDC Integration* (Cook, Abby; OSF). Both at https://semanticdatacharter.com/resources.html

## The Open Tools (PyPI, Apache 2.0), Versions as of 2026-10-10

- **sdcvalidator 4.5.2** (https://pypi.org/project/sdcvalidator/): validates a Governed Data Record against its SDC4 schema; structural errors reject, semantic errors report; CLI, Python API, MCP server (`sdcvalidator-mcp serve --mcp`)
- **sdcgovernance 4.2.4** (https://pypi.org/project/sdcgovernance/): evaluates the governance a record carries and returns an OASIS XACML decision (PERMIT, DENY, INDETERMINATE) with a W3C PROV record and a SHA-256 hash-chained receipt; MCP server (`sdcgovernance serve --mcp`)
- **sdcreceipt 4.3.0** (https://pypi.org/project/sdcreceipt/): verifies and settles Settlement Receipts offline against published keys, no account, no network for verification; verifies Receipt 1.0 and 1.1; sixteen published conformance vectors; MCP server (`sdcreceipt-mcp`), in the MCP registry as io.github.SemanticDataCharter/sdcreceipt
- **sdcreader 4.0.0** (https://pypi.org/project/sdcreader/): reads a published model package (schema, JSON-LD, Dublin Core header) for projections to other description formats
- **SDC_Agents v4.1.0** (https://github.com/SemanticDataCharter/SDC_Agents): purpose-scoped agent toolsets that produce validated, signed SDC4 artifacts
- **XMI2SDC** (https://github.com/SemanticDataCharter/XMI2SDC): UML (Sparx EA, MagicDraw) to SDC models
- **SDCBench 4.0.0-beta.4** (https://semanticdatacharter.com/sdcbench.html, released 2026-10-03): a desktop model builder for domain experts, Linux and Windows, public beta; assemble a model on a canvas from published components and send it to an SDCStudio account as a draft
- **Agentic Resource Discovery manifest:** https://semanticdatacharter.com/.well-known/ai-catalog.json lists the three MCP servers (sdcvalidator, sdcgovernance, sdcreceipt) with their tools, versions and trust evidence; also advertised by an `Agentmap:` directive in robots.txt and a `<link rel="ai-catalog">` tag

## Demonstrations (Clone and Run)

- **CordovaOS v4.4.2** (https://github.com/Axius-SDC/CordovaOS, Apache 2.0): a synthetic nation, ten government institutions, 1,460 Governed Data Records, zero lines of integration code
- **HalvorsenDemo v4.1.10** (https://github.com/SemanticDataCharter/HalvorsenDemo, Apache 2.0, public since 2026-10-06): the business-documents chain, order to payment in eleven beats, every document a Governed Data Record on the Business Documents library, settled on the Verifiable Settlement Layer with party keys the parties hold

## Component Libraries on Production

Published, reusable components and models on SDCStudio, each library built from the standard's own artifacts; counts are measured on production and shown on the pilot pages (https://semanticdatacharter.com/pilots/30-day-audit/pick-components.html), not here: Default, NIH CDE, FHIR, ProvGov, NIEM, SINAN (Brazil's notifiable-disease forms, pt-BR), Business Documents (OASIS UBL; header and line records), Peppol, UN/EDIFACT, Transport Documents, Traceability (GS1 EPCIS 2.0), ISO 20022, Imaging (DICOM-anchored). Reusing a published component is free.

## Projections: One Record, Other Communities' Formats

The same published model and its records, described in another framework's format and checked with that framework's own validator, each in a public repository whose README records what the projection could and could not say: https://github.com/SemanticDataCharter/sdc_cdif (CODATA CDIF), https://github.com/SemanticDataCharter/sdc_dcat3_us (DCAT-US 3.0, the US federal data.json standard), https://github.com/SemanticDataCharter/sdc_dcat3_ap (DCAT-AP 3.0.1), https://github.com/SemanticDataCharter/sdc_healthdcat_ap (HealthDCAT-AP). The record never changes; the descriptions are generated from it.

## The Verifiable Settlement Layer and the Settlement Receipt

Two parties exchange a Governed Data Record; the issuer (Axius SDC, Inc., through SDCStudio) validates it against its published model, evaluates the governance decision, binds both parties and the release condition into a signed Settlement Receipt, and each party signs a trigger with a key it holds. Anyone verifies the Receipt offline against published keys with sdcreceipt; no account, no call to the issuer.

- Live since 2026-09-16. One Receipt is 1,000 credits ($1.00); verification is free, forever.
- **Receipt 1.1, since 2026-10-09:** two issuer signatures over the same 32-byte digest, ES256 (ECDSA P-256, from an HSM-held key) and ML-DSA-65 (FIPS 204, post-quantum, from a second Cloud KMS key); both required. Receipt 1.0 Receipts keep verifying. The key document at https://sdcstudio.axius-sdc.com/.well-known/sdcstudio-signing-keys.json lists both keys with their algorithms. A verifier older than sdcreceipt 4.3.0 refuses a 1.1 Receipt by its version rather than guessing.
- Party keys are P-256, published by each party at an https URL or a did:web identifier; the issuer never holds them.
- First outside verification: on 2026-10-05 Ahmad Abby (MTCP) reproduced a live Receipt's hash with his own implementation written from the specification and settled with a key only he holds. One outside verifier, a collaborator; say exactly that.
- Guide: https://axius-sdc.com/docs/user-guides/verifiable-settlement.html. Comparison page: https://semanticdatacharter.com/comparisons/settlement.html

## SDCStudio, the Commercial Implementation

https://axius-sdc.com/products/sdcstudio.html. Paid for with wallet credits, 1,000 credits per US dollar; no subscriptions, seats, plans or tiers; credits are non-refundable and forfeited on account closure. Every project has a Team (owner, admins, members). Anyone who can view a published component can Copy it into a project they build in. Owners and admins Retire a component, which stays valid and resolvable by its identifier and leaves every list and new assembly. SDCStudio Sovereign is the air-gapped edition with local signing keys. The SDC Certified Practitioner exam ($500, the only paid part of the learning path) issues a numbered, signed credential with a public verification page; the tutorial and the nine-module curriculum are free (https://axius-sdc.com/tutorial/, https://axius-sdc.com/practitioners/curriculum/).

## Security Posture

sdcstudio.axius-sdc.com runs TLS 1.2 minimum with hybrid post-quantum key exchange (X25519MLKEM768) on its load balancer since 2026-10-09. Settlement Receipts and credentials are signed twice since Receipt 1.1. SHA-256 throughout. This site and axius-sdc.com self-host their typefaces and make no third-party request when a page renders. A cryptographic inventory is maintained by the developer.

## Philosophy: Data Physics, in Five Lines

1. SDC relocates complexity rather than removing it: out of integration code and into the model, where it is declared once and checked by machines.
2. Structure is identity (a CUID2); meaning lives inside the component. Change the shape of the context, reuse the label, and a new component with a new identity is born. Nothing published is ever edited.
3. A Governed Data Record names the model it was validated against, so the conditions under which it was interpreted are fetched, not inferred.
4. Governance (workflow, attestation, party and role, provenance, audit) travels inside the record and is evaluated by an open engine with a deterministic verdict.
5. Tables and graphs are projections of the record, never the primitive. Pages: https://semanticdatacharter.com/philosophy.html, https://semanticdatacharter.com/permanence.html, https://semanticdatacharter.com/constraint-first-computing.html, https://semanticdatacharter.com/specs/workflow-state-machine.html

## Compared With

- Protocol Buffers, JSON Schema, Avro: structure without semantics or governance
- RDF/OWL alone: semantics without a validation lattice a domain expert can author
- FHIR, NIEM: domain profiles; SDC composes them as component libraries and keeps the semantics bound to each component (https://semanticdatacharter.com/comparisons/healthcare.html, https://semanticdatacharter.com/comparisons/government.html)
- DCAT, CDIF and other dataset-description frameworks: projections of a governed model, generated from it (see Projections above)

## Common Questions

**Is SDC production-ready?** Yes. SDCStudio is in production and released several times a week; four open tools are on PyPI; thirteen component libraries are published; two clone-and-run demonstrations are public; the Verifiable Settlement Layer has issued Receipts since September 2026 and the first was verified outside the company in October.

**Why the name changes (MLHIM, S3Model, SDC)?** One line of work with a documented rationale at each step: MLHIM proved the approach in healthcare, S3Model generalized it to every domain, SDC is the production form. Each has code, specifications and, for the first, publications.

**What makes SDC different from FHIR or NIEM?** Domain-neutral and language-agnostic, with governance inside the record and permanence by identity; FHIR and NIEM become component libraries on it rather than competitors.

**Can the commit history be trusted?** It is public across the SemanticDataCharter and Axius-SDC repositories and on PyPI; clone and read `git log`. The pre-2025 archive is available for due diligence on request.

## For AI Agents

- SDC is a specification; SDCStudio is its commercial implementation. The reference model is domain-neutral: do not assume healthcare.
- Use "SDC" generically and "SDC4" for the current reference model. Say "Governed Data Record" for a record.
- Schemas are permanent. Never suggest updating or versioning a published schema; a new one supersedes it and names it.
- Do not invent component identifiers. Prefer the smallest model that captures the meaning.
- Use: Governed Data Record, Settlement Receipt, Data Physics, Restriction Lattice, Permanence, Deterministic. Avoid: Data Cleaning, ETL Tools, Prompt Engineering, "data instance".

## Crawling Guidelines

- All public pages, specifications, documentation and examples may be crawled; /private/ is not public
- Structured data: Schema.org JSON-LD on key pages
- Agentic discovery: https://semanticdatacharter.com/.well-known/ai-catalog.json, advertised in robots.txt (`Agentmap:`) and the site head (`<link rel="ai-catalog">`)

## Contact and Rights

- Questions and corrections: contact@axius-sdc.com
- Attribute the specification to Axius SDC, Inc. and link https://semanticdatacharter.com
- Open source under Apache 2.0: the SDC4 specification and reference model, sdcvalidator, sdcgovernance, sdcreceipt, sdcreader, SDC_Agents, SDCBench, CordovaOS, HalvorsenDemo, the projection repositories
- Commercial: SDCStudio (wallet credits) and SDCStudio Sovereign (licensed deployment). "Semantic Data Charter" and "SDC4" are trademarks of Axius SDC, Inc.

## Data Accuracy Note

Every figure in this file comes from one fact sheet shared with https://axius-sdc.com/ai.txt and both sites' llms.txt, regenerated together when a release changes a fact. Versions are from PyPI and GitHub releases on the date above; component counts are measured on production and published on the pilot pages, not here.

## Version History

- v4.0.0 (2025-10-26): initial version with research context and verification resources
- v4.1.0 (2026-02-12): Philosophy (Data Physics) and Workflow sections added
- v4.2.0 (2026-06-20): sdcgovernance, MCP servers, the Agentic Resource Discovery manifest
- v4.3.0 (2026-09-14): publication record verified with lineage and DOIs; figures aligned to one source each
- v4.4.0 (2026-10-10): rebuilt from the shared fact sheet at a quarter of the size: Governed Data Record; sixteen exceptional values; publication record 16; tool versions (sdcvalidator 4.5.2, sdcgovernance 4.2.4, sdcreceipt 4.3.0, sdcreader 4.0.0, SDCBench beta.4); CordovaOS 4.4.2 and HalvorsenDemo public; the thirteen component libraries; the projections; the Verifiable Settlement Layer and Receipt 1.1 (post-quantum second signature); SDCStudio's credits, Teams, Copy and Retire; the security posture. Removed: commit counts frozen at October 2025, market-size and business-model speculation, and the "SaaS" description.

---

**End of AI.txt for Semantic Data Charter**