Standards & Specifications

Open protocols for privacy-preserving identity, human communication, and consent-aware AI. Specifications are open source. Patent licensing terms vary by family. See licensing for details.

Three things that do not exist yet

Identity that evolves

Current digital identity is static: a username, a certificate, a credential. Your real identity is not static. You gain qualifications, change employers, move countries, build relationships. Your identity is a constellation of weighted connections that shifts over time.

NUID models identity as a weighted graph. When you get a new attestation, your constellation gains a node. When a credential expires, the weight decays. When you move jurisdictions, the regulatory context updates automatically. Your identity is always current, always yours, and works the same over internet, mesh, Bluetooth, or offline.

Consent that propagates

Today, consent is a checkbox. You click "I agree" and your data flows everywhere with no way to trace it or take it back. NOEVA consent is cryptographic. It has a scope, an expiry, and it propagates through every system that touches your data.

If you revoke consent, that revocation propagates to every downstream system. Not because they choose to respect it, but because the protocol makes it impossible to process data against a revoked consent token. Consent always expires. It cannot be made perpetual. This is enforced at the protocol level.

Content with provenance

When an AI generates a summary of your document, who wrote it? When that summary gets shared, edited, and reshared, what came from you and what was added later? Today there is no answer. Content provenance solves this.

Every piece of content carries a cryptographic chain showing who created it, how it was transformed, and what consent governs its use. If an AI adapts content for a different audience, the provenance chain records the transformation. The original creator's intent and rights travel with the content, wherever it goes.

The protocols

NUID

One identity that is yours, works everywhere, and cannot be taken away.

W3C DID WG

HCP

A way for people to communicate that works for everyone, including those with disabilities.

IETF Draft

CRL

How multiple AI systems work together without any of them being able to bypass safety.

NOEVA Spec

FAPA

Proving who you are without a central authority keeping a list.

DIF

AHP

How AI agents and humans talk to each other with clear rules both sides can trust.

NOEVA Spec

ACP

Content that adapts to whoever is reading it, in whatever format they need.

W3C

Content Provenance

Where content came from, how it was transformed, and who has rights over it.

Families 24, 30-31

NUID

W3C DID WG

NOEVA Universal Identifier, did:nuid Method Specification

One identity that is yours, works everywhere, and cannot be taken away. Your NUID carries your jurisdiction, your entity type, and your temporal scope, so any system that sees it knows exactly what rules apply, without looking anything up.

Identifier Structure

did:nuid:<jurisdiction>:<entity-type>:<epoch>:<unique-id>

Examples:
did:nuid:uk:person:2026:abc123def456     // UK individual
did:nuid:eu:org:2026:acme-corp           // EU organisation
did:nuid:self:device:2026:iphone-x1      // Self-controlled device
did:nuid:global:service:2026:resolve-ai  // Global service

Jurisdiction Scopes

Scope	Meaning	Jurisdiction
self	Self-controlled	None (individual control)
uk	UK jurisdiction	UK GDPR, Data Protection Act 2018
eu	EU jurisdiction	GDPR, AI Act, eIDAS
global	Cross-jurisdictional	Highest common standard

Entity Types

person org device service agent collective

The epoch component enables temporal scoping for consent validity and key rotation without breaking identifier stability.

The Constellation Model

A NUID is not just a string. It is the anchor of a constellation: a weighted graph of attestations, relationships, and context that evolves with you.

Attestations as nodes. Your university degree, your employer verification, your age proof: each is a weighted node in your constellation, issued by a different authority.

Weights decay over time. A credential from last week carries more weight than one from five years ago. Relevance is temporal, and the protocol reflects that.

Context-dependent disclosure. When you apply for a job, your constellation presents employment-relevant nodes. When you cross a border, it presents jurisdiction-relevant ones. You control which parts of your constellation are visible in each context.

Transport-agnostic. The same constellation works over internet, mesh, Bluetooth, NFC, or physical media. Your identity does not depend on connectivity.

SDK → · Spec → · Mesh Identity →

HCP

IETF Draft

Human Communication Protocol

A way for people to communicate that works for everyone, including those with disabilities. Accessibility is a protocol-level requirement, not an afterthought. If a message is not accessible, it does not send.

6 Mandatory Blocks

Every HCP message contains exactly 6 blocks. Missing blocks cause protocol rejection:

1. Identity Block Sender/receiver NUIDs, channel binding

2. Safety Block Content warnings, consent requirements, age gates

3. Accessibility Block Alt text, captions, reading level, language

4. Meaning Block Semantic intent, emotion, context

5. Content Block Actual message content (text, media, data)

6. Routing Block Delivery preferences, retention, deletion proofs

Accessibility Enforcement

If a message contains an image without alt text, it is rejected at the protocol level.

Cryptographic Deletion

HCP includes verifiable deletion proofs with witness signatures, ensuring data can be provably removed.

GitHub →

CRL

NOEVA

Cognitive Routing Layer

How multiple AI systems work together without any of them being able to bypass safety. When no safe route exists, the system refuses to act. That is designed behaviour, not failure.

Core Principle: Refusal-to-Route

if no_safe_path_exists(request, constraints):
    return CRL_NO_SAFE_PATH  // Designed behaviour, not failure

Criticality Tiers

Tier	Description	Routing Requirement
Low	Non-sensitive operations	Single model sufficient
Medium	Standard sensitivity	Verified model required
High	Sensitive data/decisions	Multiple model agreement
Critical	Life-affecting decisions	Cross-vendor consensus + human oversight

Failure Taxonomy

CRL_NO_SAFE_PATH CRL_TIMEOUT CRL_REFUSAL CRL_CONTRADICTION CRL_SAFETY_VIOLATION

EU AI Act Alignment

CRL directly supports High-Risk AI requirements: risk management via criticality tiers, human oversight for Critical tier, accuracy through cross-vendor consensus, complete audit trails.

GitHub →

FAPA

DIF

Federated Attestation Protocol Architecture

Proving who you are without a central authority keeping a list. Different organisations can vouch for different things about you, and those attestations combine into a picture of trust that no single entity controls.

Key Features

Multi-issuer attestation chains. Combine attestations from your bank, your employer, your university, and your government into a single trust picture, without any of them talking to each other.

Selective disclosure. Prove you are over 18 without revealing your date of birth. Prove you have a degree without revealing your transcript.

Cross-jurisdiction trust bridging. An attestation from a UK institution can be verified in the EU, with both jurisdictions' rules respected automatically.

Revocation with privacy. Revoke an attestation without revealing to third parties that it was revoked, or why.

AHP

NOEVA

Agent Human Protocol

How AI agents and humans talk to each other with clear rules both sides can trust. Every interaction respects your jurisdiction, your privacy settings, and your consent boundaries, automatically.

Transport Layer

stdio with NDJSON (Newline-Delimited JSON) for low-latency communication:

{"jsonrpc":"2.0","method":"session/create","params":{"nuid":"did:nuid:self:person:2026:dev1"},"id":1}
{"jsonrpc":"2.0","method":"agent/invoke","params":{"capability":"code_review","context":{}},"id":2}

Session Management

Sessions maintain context across interactions with automatic privacy constraint propagation. All routing decisions respect the user's NUID jurisdiction scope.

ACP

W3C

Adaptive Content Protocol

Content that adapts to whoever is reading it, in whatever format they need. A complex legal document can become a simple summary. A visual chart can become a spoken description. The meaning stays the same; the format fits the person.

Semantic Core

Every piece of content has a semantic core that must be preserved through transformation:

interface SemanticCore {
    meaningGraph: MeaningNode[];      // What the content means
    invariants: Invariant[];           // What must NOT change
    adaptationLimits: Limits;          // Boundaries for transformation
    creatorIntent: Intent;             // Original purpose
}

Transformation Pipeline

Content flows through: Parse → Extract Semantic Core → Apply Viewer Preferences → Transform → Verify Invariants → Deliver

Content Provenance

Families 24, 30-31

Provenance Chain, Credibility Signals, and Transformation Tracking

Every piece of content carries a cryptographic chain showing where it came from, how it was transformed, and what consent governs its use. When AI transforms content, the chain records exactly what changed.

Provenance Chain

ContentNode {
  creator: NUID              // Who created this
  created: Timestamp          // When
  consent: ConsentToken       // Under what terms
  transformations: Chain[]    // Every change, in order
  semanticHash: Hash          // Meaning fingerprint
}

Transformation {
  agent: NUID                 // Who/what made the change
  type: "summarise" | "translate" | "adapt" | ...
  preservedInvariants: Hash   // Proof meaning was preserved
  parentHash: Hash            // Links to previous version
}

What This Enables

Attribution that survives transformation. When an AI summarises your article, the summary links back to you. When someone shares that summary, your attribution travels with it.

Consent that travels with content. If you share a document under specific terms, those terms are encoded in the provenance chain. Downstream systems can verify what is permitted without contacting you.

Credibility signals. How many transformations has this content been through? Were they by verified agents? Does the semantic hash still match the original meaning? These are computable questions, not opinions.

Revocation propagation. If you revoke consent for your content, the revocation reaches every copy because each copy carries the consent token that is now invalid.

SDK → · Spec → · Consent Protocol →

Licensing

NOEVA specifications are released under:

Apache 2.0 for code and reference implementations

CC BY-SA 4.0 for documentation

Core protocol patents are pledged open under RAND-Z terms for open source implementations

Commercial patents require a separate licence

See the licensing page for the full patent pledge and licensing options.