Open standard · Technical specification

EntityMap specification

A structured, entity-first index of website content for AI agent and LLM consumption

Version 1.0

Status Stable

Date 2026-04-07

Author Fred Laurent

License CC BY 4.0

Files entitymap.json · entitymap.html

Contents

Abstract
Conformance floor - minimum valid file
File conventions
JSON structure
Entity types
Predicate vocabulary
The HTML companion file
Validation
Appendix A - Minimal valid example
Appendix B - Predicate reference
Appendix C - Consumer conformance levels
Appendix D - Extension profiles
Appendix E - Changelog

-Abstract

EntityMap is an open standard for publishing a structured, entity-first index of a website's content, designed for consumption by AI agents, large language models, and RAG pipelines.

Where sitemap.xml tells crawlers what pages exist, entitymap.json tells AI systems what a site knows - which entities it covers, how they relate, and where the evidence is.

EntityMap v1.0 is a publisher assertion standard. It requires a small mandatory core - roughly 12 fields across three objects. Everything beyond that core is optional enrichment that improves reasoning quality, attribution, and graph depth without affecting basic conformance.

The key words MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY are interpreted as described in RFC 2119. Publisher - the organisation operating the described website. Entity - a named concept, person, product, or other identifiable thing covered by the publisher's content. Chunk - an extractive passage from a source page, associated with one entity. Consumer - any AI agent, LLM, RAG pipeline, or crawler that reads an EntityMap.

1.Conformance floor - minimum valid file

A conforming EntityMap v1.0 file requires exactly three things: a valid root object, at least one entity object, and at least one chunk per entity. Everything else in this specification is optional enrichment.

Minimum valid entitymap.json

{
  "version": "1.0",
  "schema": "https://entitymap.org/spec/v1.0",
  "publisher": {
    "name": "Acme Corp",
    "url": "https://acme.com"
  },
  "generated": "2026-04-07T00:00:00Z",
  "entities": [
    {
      "entityId": "e_001",
      "@type": "Concept",
      "name": "Companion Planting",
      "description": "The practice of growing different plants
        in proximity for mutual benefit.",
      "hasChunks": [
        {
          "chunkId": "c_001",
          "text": "Companion planting pairs plants that benefit each other.",
          "sourceUrl": "https://acme.com/companion-planting",
          "pageTitle": "Companion Planting Guide",
          "publisher": "Acme Corp"
        }
      ]
    }
  ]
}

Object	Required fields
Root	`version`, `schema`, `publisher.name`, `publisher.url`, `generated`, `entities`
Entity	`entityId`, `@type`, `name`, `description`, `hasChunks` (min 1)
Chunk	`chunkId`, `text`, `sourceUrl`, `pageTitle`, `publisher`
Relation (if used)	`predicate`, `targetName` - plus `confidence` if predicate is Tier 3

Two hard rules apply at all enrichment levels:

publisher on every chunk MUST exactly match publisher.name in the root - including case and spacing. This is the attribution mechanism.
Tier 3 predicates (IMPROVES, DEGRADES, LEADS_TO, SUITED_FOR, TARGETS, ACHIEVES) MUST carry a confidence field.

2.File conventions

Both files MUST be served from the root of the domain without authentication.

File	URL	Purpose
`entitymap.json`	`https://example.com/entitymap.json`	Machine-readable primary file
`entitymap.html`	`https://example.com/entitymap.html`	Crawler and human-readable view

Discovery

Declare the EntityMap via robots.txt (proposed convention), a <link> tag in every page's <head>, and a sitewide footer link - the most reliable mechanism for crawlers:

# robots.txt
EntityMap: https://example.com/entitymap.json

<!-- <head> -->
<link rel="entitymap" type="application/json" href="https://example.com/entitymap.json" />

<!-- footer -->
<a href="https://example.com/entitymap.html">EntityMap</a>

Publishers SHOULD also list entitymap.html in sitemap.xml with priority: 0.9 and changefreq: weekly. entitymap.html MUST NOT carry a noindex directive.

Large sites - sharding

For generated EntityMaps exceeding 200 entities, the EntityMap SHOULD be sharded. Sharding is a transport concern - split by size, not by entity type. The root entitymap.json acts as a manifest listing all shards with their entity counts and lastModified timestamps. Each shard file MUST carry a shardOf field pointing back to the root manifest URI. Consumers load all shards - the split carries no semantic meaning.

/entitymap.json              ← manifest
/entitymap/part-001.json
/entitymap/part-002.json

3.JSON structure

3.1 Root object

{
  "version": "1.0",
  "schema": "https://entitymap.org/spec/v1.0",
  "publisher": { ... },
  "generated": "2026-04-07T00:00:00Z",
  "entities": [ ... ],

  "profile": "core",
  "verificationStatus": "third-party-verified",
  "certification": { ... },
  "previousVersion": "https://...",
  "changeLog": [ ... ],
  "shards": [ ... ],
  "vocabulary": { ... }
}

Field	Conformance	Description
`version`	MUST	Must be "1.0".
`schema`	MUST	Must be "https://entitymap.org/spec/v1.0".
`publisher`	MUST	Publisher identity object. See §3.1 publisher fields below.
`generated`	MUST	ISO 8601 timestamp. MUST be updated on every rebuild.
`entities`	MUST	Array of entity objects. Min 1.
`profile`	MAY	Extension profile. Default: "core". See Appendix D.
`verificationStatus`	MAY	Trust level declared by the publisher. Allowed values: `"self-declared"` / `"generator-draft"` / `"third-party-verified"`. Default: `"self-declared"`. SHOULD be set to `"third-party-verified"` when a valid `certification` field is present. Consuming tools MUST treat this field as a hint - the `certification` registry is the authority. See §3.6.
`certification`	MAY	Third-party certification object issued by entitymap.org. Presence does not imply validity - tools MUST verify against the live registry. See §3.6.
`previousVersion`	MAY	URI of prior entitymap.json. Enables consumer diffing.
`changeLog`	MAY	Array of change entries (added / deprecated / modified / merged). deprecated and merged entries MUST include `replacedBy`.
`shards`	MAY	Index of shard files with entity counts and `lastModified` timestamps. See §2.
`vocabulary`	MAY	Custom predicate declarations. See §6.4.

Publisher fields

Field	Conformance	Description
`name`	MUST	Canonical brand name. MUST NOT be a domain, product name, or generic descriptor. MUST match `publisher` on all chunks exactly.
`url`	MUST	Canonical URL of the publisher.
`sameAs`	MAY	Wikidata or Wikipedia URI anchoring publisher to the open knowledge graph.

3.2 Entity object

{
  "entityId": "e_001",
  "@type": "ProprietaryTerm",
  "name": "AI Share of Voice",
  "description": "A metric measuring...",
  "hasChunks": [ ... ],

  "alternateName": "AI SOV",
  "canonicalLabel": "share of voice",
  "sameAs": "https://www.wikidata.org/wiki/Q...",
  "maturityStatus": "established",
  "audienceType": "technical",
  "status": "active",
  "replacedBy": null,
  "relations": [ ... ]
}

Field	Conformance	Description
`entityId`	MUST	Stable unique identifier. Never reuse a retired ID.
`@type`	MUST	v1.0 core type. See §4.
`name`	MUST	Publisher-specific label.
`description`	MUST	1–3 sentence definition as this publisher uses the concept.
`hasChunks`	MUST	1–5 evidence chunks. See §3.4.
`replacedBy`	MUST if deprecated/merged	entityId of the replacement entity.
`sameAs`	SHOULD for Concept	Wikidata or Wikipedia URI. Strongly recommended for Concept type; optional for others.
`alternateName`	MAY	Abbreviation or surface form variant. Aids disambiguation.
`canonicalLabel`	MAY	General concept label where the publisher uses a proprietary variant.
`maturityStatus`	MAY	proposed / established / deprecated.
`audienceType`	MAY	technical / executive / general / regulatory.
`status`	MAY	active / deprecated / merged. Default: active.
`relations`	MAY	Typed relationships to other entities. See §3.3.

3.3 Relation object

{
  "predicate": "IMPROVES",
  "targetId": "e_004",
  "targetName": "Retrieval Precision",
  "confidence": "declared",

  "context": {
    "condition": "when chunks are publisher-attributed",
    "temporal": "2024-onwards",
    "jurisdiction": null,
    "reviewedBy": "Fred Laurent",
    "reviewDate": "2026-04-01"
  },
  "targetUri": "...",
  "targetShard": "...",
  "targetDescription": "..."
}

Field	Conformance	Description
`predicate`	MUST	From standard vocabulary (§6) or declared custom vocabulary (§6.4).
`targetName`	MUST	Human-readable target name. Required in all cases - survives aggregation.
`confidence`	MUST for Tier 3	declared / inferred. Required on Tier 3 predicates; optional on Tier 1/2.
`targetId`	SHOULD	entityId of internal target.
`context`	MAY	Qualification object: condition, temporal, jurisdiction, reviewedBy, reviewDate.
`targetUri`	MAY	URI for external entities (Wikidata, Wikipedia, schema.org).
`targetShard`	MAY	Path to shard file containing target entity.
`targetDescription`	MAY	One-sentence summary of target. SHOULD be present when targetUri is absent.

A confidence: "inferred" relation without a context object will produce a validator warning - consuming systems discount heavily without qualification context.

3.4 Chunk object

{
  "chunkId": "c_001",
  "text": "...",
  "sourceUrl": "https://acme.com/page",
  "pageTitle": "Page Title",
  "publisher": "Acme Corp",

  "retrieved": "2026-04-07T09:00:00Z",
  "relevanceScore": 0.92,
  "contentType": "definition",
  "audienceType": "technical"
}

Field	Conformance	Description
`chunkId`	MUST	Unique identifier within this EntityMap.
`text`	MUST	Evidence passage. 1–5 sentences, max 600 characters. SHOULD be extractive.
`sourceUrl`	MUST	Canonical URL of source page. MUST be publicly accessible.
`pageTitle`	MUST	Title of source page at time of retrieval.
`publisher`	MUST	MUST exactly match `publisher.name` in root - including case and spacing.
`retrieved`	SHOULD	ISO 8601 timestamp when the source was fetched.
`relevanceScore`	MAY	Float 0.0–1.0. Publisher-assigned relevance to its entity.
`contentType`	MAY	definition / evidence / example / statistic / procedure.
`audienceType`	MAY	technical / executive / general / regulatory.

3.5 Attribution requirements

This section is normative.

Publisher identity. publisher.name MUST be a canonical brand name - not a domain, product name, or generic descriptor. It is the name that will appear in AI-generated attribution.

Chunk-level attribution. The publisher field on every chunk MUST exactly match publisher.name. Chunks are extracted and stored independently in vector databases - the publisher field is the mechanism by which attribution survives that extraction. Case differences, abbreviations, and trailing whitespace all constitute a mismatch.

Freshness. generated MUST be updated on every rebuild. A timestamp older than 30 days signals potential staleness to consumers.

Canonical labelling. Where a publisher uses a proprietary term for a widely-known concept, canonicalLabel carries the general term while name carries the publisher-specific term. This aids cross-publisher disambiguation without losing the publisher's terminology.

3.6 Certification and verification status

This section is normative.

EntityMap provides two complementary trust signals at the root level: verificationStatus, which is publisher-declared, and certification, which is issued by a third-party registry. The registry is the authority - verificationStatus in the file is a hint that consuming tools MUST NOT treat as a guarantee.

The certification object

{
  "certification": {
    "url": "https://entitymap.org/certified/acme.com/a3f8c2d1e9b47f6c8d2e1a9b3f7c4d8e",
    "issuedAt": "2026-04-21T09:00:00Z",
    "expiresAt": "2026-07-20T09:00:00Z"
  }
}

Field	Conformance	Description
`url`	MUST if object present	Registry URL in the form `https://entitymap.org/certified/{domain}/{token}`. The `{domain}` segment MUST exactly match the hostname of `publisher.url` (without scheme or trailing slash). `{token}` is a 32-character lowercase hex string. `GET` returns `200` (certified) or `404` (not certified, expired, or revoked).
`issuedAt`	SHOULD	ISO 8601 timestamp of when this certification was issued.
`expiresAt`	SHOULD	ISO 8601 timestamp of expiry. Certifications expire after 90 days. Tools MAY warn publishers within 14 days of expiry but MUST NOT downgrade certified status before actual expiry.

verificationStatus values

Value	Meaning	Typical context
`"self-declared"`	Publisher asserts accuracy. No third-party verification.	Default for hand-written or manually reviewed entitymaps.
`"generator-draft"`	Produced by an automated generator without human review. Consumers SHOULD apply lower reasoning weight to Tier 3 predicates.	Output of any automated generation pipeline prior to publisher review.
`"third-party-verified"`	Publisher claims third-party certification. MUST be backed by a valid `certification` field. Without one, treat as `"self-declared"`.	Set after receiving a valid certification token from entitymap.org.

Normative consuming rules

certification.url present	Registry response	Tool MUST treat as
Yes	`200`	`third-party-verified` - regardless of declared verificationStatus.
Yes	`404`	`self-declared` - cert expired or revoked. Tool MAY surface a warning to the publisher.
Yes	Unreachable	Unknown. Use cached status (max 24h) if available. Do not assume either state.
No	-	Trust `verificationStatus` as declared. If declared `"third-party-verified"` without a `certification` field, treat as `"self-declared"`.

Domain binding

The {domain} segment of certification.url MUST match the hostname of publisher.url. Consuming tools MUST verify this before making a registry request. A mismatch indicates a token copied from another domain and MUST be treated as uncertified without contacting the registry.

// Domain binding check (pseudocode)
certDomain    = extract_hostname(certification.url)   // "acme.com"
publisherHost = extract_hostname(publisher.url)        // "acme.com"

if certDomain !== publisherHost → treat as uncertified, skip registry call

A certification token is bound to a single domain at issuance. A token issued for acme.com produces registry URLs of the form entitymap.org/certified/acme.com/{token}. Using that token on any other domain - including subdomains - produces a URL the registry does not recognise, returning 404.

Publisher obligations

A publisher holding a valid certification SHOULD keep verificationStatus set to "third-party-verified" and SHOULD update certification.expiresAt on renewal. On expiry or revocation, publishers SHOULD either renew or remove the certification field and revert verificationStatus to "self-declared". Leaving an expired certification field in place is not a spec error - consuming tools handle it correctly via the registry check - but it is misleading to human readers of the file.

The certification registry and submission process will be available at entitymap.org/certification. Publishers MAY include the certification field in files now - the field is fully specified and validator-checked. The live registry launches Q3 2026.

4.Entity types

EntityMap v1.0 defines 15 core types in three tiers reflecting the epistemic role of the entity. Publishers MUST use a v1.0 core type or a namespaced custom type (e.g. "acme:MetricComponent").

Tier 1 - Knowledge

Tier 1 · Knowledge

Concept

General domain term. Common knowledge. Add sameAs. Consumers blend with general priors.

Tier 1 · Knowledge

ProprietaryTerm

Publisher-coined concept. Definition here is authoritative. No sameAs expected.

Tier 1 · Knowledge

Methodology

Named process, framework, or approach.

Tier 1 · Knowledge

Metric

Measurable quantity with defined calculation. Source of MEASURES relations.

Tier 1 · Knowledge

Taxonomy

Classification system the publisher maintains. Use COVERS for sub-categories.

Tier 2 - Actor

Tier 2 · Actor

Person

Named individual. Use AFFILIATED_WITH for their organisation.

Tier 2 · Actor

Organization

Company, institution, or body.

Tier 2 · Actor

SoftwareProduct

Software application, SaaS tool, API, or developer platform.

Tier 2 · Actor

PhysicalProduct

Tangible goods.

Tier 2 · Actor

Service

Professional or subscription offering. Not software.

Tier 2 · Actor

Platform

Multi-sided or ecosystem-enabling product.

Tier 2 · Actor

Place

Geographic location or venue the publisher has content authority over. Add sameAs to Wikidata, Wikipedia or Geonames.

Tier 3 - Temporal

Tier 3 · Temporal

Event

Named occurrence with a defined time.

Tier 3 · Temporal

Standard

Specification or protocol with a version and governance body.

Tier 3 · Temporal

Regulation

Formal legal or regulatory instrument. Target of REGULATED_BY.

Type decision rules

Concept vs ProprietaryTerm: Does this concept exist independently of the publisher? → Concept with sameAs. Did the publisher coin or materially define it? → ProprietaryTerm.

SoftwareProduct vs Platform vs Service: Primarily software? → SoftwareProduct. Ecosystem or developer layer is central? → Platform. Primarily human-delivered? → Service.

Standard vs Regulation: Formally enacted into law? → Regulation. Voluntary specification with governance body? → Standard.

5.Predicate vocabulary

All predicates are uppercase. Three tiers by semantic hardness determine the confidence field requirement and consumer trust behaviour. Full definitions and examples: entitymap.org/predicates.

Tier 1 - Hard predicates (11)

Unambiguous, machine-trustable. No confidence field required. Inverses are implicit - never declare both directions of PART_OF/INCLUDES.

INSTANCE_OF

PART_OF

INCLUDES

DEPENDS_ON

REQUIRES

MEASURES

PRODUCED_BY

REGULATED_BY

AUTHORED_BY

AFFILIATED_WITH

COVERS

MEASURES: source must be Metric · AFFILIATED_WITH: source must be Person · COVERS: source must be Concept, ProprietaryTerm, or Taxonomy

Tier 2 - Structural predicates (7)

Clear semantics; directional discipline required. confidence optional. RELATES_TO is the predicate of last resort - use only when no other predicate fits.

RELATES_TO

PRECEDES

ENABLES

PREVENTS

CONFLICTS_WITH

DESCRIBED_BY

OFFERS

Tier 3 - Interpretive predicates (6)

Carry editorial judgment. confidence is required - validator errors if absent. Consumers apply lower reasoning weight when confidence: "inferred".

IMPROVES

DEGRADES

LEADS_TO

SUITED_FOR

TARGETS

ACHIEVES

Predicate decision rules

PART_OF vs DEPENDS_ON: Definitional constituent → PART_OF. Separate concept needing the other to function → DEPENDS_ON.

INCLUDES vs COVERS: Object is a component of subject → INCLUDES. Subject is a hub and object is a sub-topic the publisher covers → COVERS.

ENABLES vs IMPROVES: Structural enablement, unambiguous → ENABLES (Tier 2). Causal effect requiring editorial judgment → IMPROVES (Tier 3, confidence required).

TARGETS vs SUITED_FOR: Designed for the object → TARGETS. Happens to fit well but not designed for it → SUITED_FOR.

Custom predicates

"vocabulary": {
  "predicates": ["POLLINATES", "ZONES_AS"],
  "namespace": "https://acme.com/entitymap/vocab/v1"
}

Custom predicates MUST be uppercase, MUST NOT conflict with standard names, and MUST be documented at the declared namespace URI.

6.The HTML companion file

entitymap.html is generated from entitymap.json and MUST NOT be maintained independently. A conforming entitymap.html MUST:

Reference entitymap.json via <link rel="alternate" type="application/json">
Embed per-entity JSON-LD in <script type="application/ld+json"> blocks
Render relations as internal hyperlinks where targets exist in the same file
Include a data-publisher attribute on every chunk blockquote
Render the publisher name as visible plain text in every chunk's <cite> element - pattern: [page title] - published by [publisher name]
Not carry a noindex directive

The visible-text attribution requirement exists because many LLM pipelines strip HTML tags before ingestion, discarding all metadata. Publisher attribution that lives only in structured attributes is invisible to those systems. The cite text is the fallback that survives plain-text ingestion.

<blockquote data-publisher="Acme Corp">
  "Chunk text here."
  <cite>
    <a href="https://acme.com/page">Page title</a> - published by Acme Corp
  </cite>
</blockquote>

7.Validation

A validator is available at entitymap.org/validate. The following conditions produce errors (not warnings):

Missing any required field at root, entity, or chunk level
Using a v0.x legacy type
Chunk publisher not exactly matching publisher.name
Tier 3 predicate without a confidence field
Entity with status: "deprecated" or "merged" without replacedBy
MEASURES used on a non-Metric source entity
AFFILIATED_WITH used on a non-Person source entity
COVERS used on a non-hub source type
Both directions of PART_OF/INCLUDES declared between the same entity pair
IMPROVES and DEGRADES, or ENABLES and PREVENTS, declared between the same entity pair
Chunk text exceeding 600 characters
More than 5 chunks per entity
certification.url present but domain segment does not match publisher.url hostname
certification.url present but not in the form https://entitymap.org/certified/{domain}/{token}

The validator also produces advisory warnings for recommended improvements beyond the mandatory floor, including missing sameAs on Concept types, overuse of RELATES_TO, and verificationStatus: "third-party-verified" declared without a certification field.

LLM-assisted generators produce draft EntityMaps. Files produced without human review MUST be published with verificationStatus: "generator-draft". The confidence: "declared" designation and the ProprietaryTerm type require explicit human review. Reference implementation: waikay.io/entitymap.

A.Appendix A - Minimal valid example

{
  "version": "1.0",
  "schema": "https://entitymap.org/spec/v1.0",
  "publisher": {
    "name": "Acme Gardens",
    "url": "https://acmegardens.com"
  },
  "generated": "2026-04-07T00:00:00Z",
  "entities": [
    {
      "entityId": "e_001",
      "@type": "Concept",
      "name": "Companion Planting",
      "description": "The practice of growing different plants in proximity
        for mutual benefit, including pest control, pollination support,
        and improved yield.",
      "sameAs": "https://www.wikidata.org/wiki/Q905413",
      "relations": [
        {
          "predicate": "IMPROVES",
          "targetId": "e_002",
          "targetName": "Crop Yield",
          "confidence": "declared"
        },
        {
          "predicate": "PREVENTS",
          "targetId": "e_003",
          "targetName": "Pest Damage"
        }
      ],
      "hasChunks": [
        {
          "chunkId": "c_001",
          "text": "Companion planting pairs plants that benefit each other -
            growing basil near tomatoes repels aphids and improves
            fruit flavour.",
          "sourceUrl": "https://acmegardens.com/companion-planting-guide",
          "pageTitle": "The Complete Companion Planting Guide",
          "publisher": "Acme Gardens",
          "retrieved": "2026-04-07T09:14:00Z",
          "relevanceScore": 0.95,
          "contentType": "evidence"
        }
      ]
    }
  ]
}

B.Appendix B - Predicate reference

TIER 1 - HARD (11) - no confidence required
  INSTANCE_OF     PART_OF         INCLUDES
  DEPENDS_ON      REQUIRES        MEASURES *
  PRODUCED_BY     REGULATED_BY    AUTHORED_BY
  AFFILIATED_WITH *  COVERS **

  * type-constrained source
  ** COVERS: source must be Concept, ProprietaryTerm, or Taxonomy

TIER 2 - STRUCTURAL (7) - confidence optional
  RELATES_TO †    PRECEDES        ENABLES
  PREVENTS        CONFLICTS_WITH  DESCRIBED_BY
  OFFERS

  † RELATES_TO: last resort - validator warns above 20% of all relations

TIER 3 - INTERPRETIVE (6) - confidence REQUIRED
  IMPROVES        DEGRADES        LEADS_TO
  SUITED_FOR      TARGETS         ACHIEVES

RESERVED - HEALTHCARE PROFILE (v1.1)
  TREATS  CONTRAINDICATED_WITH  REDUCES  INDICATES  EVIDENCED_BY

RESERVED - FINANCE PROFILE (v1.1)
  CORRELATED_WITH  BENCHMARKS_AGAINST  PRICED_BY  HEDGES

RESERVED - EDUCATION PROFILE (v1.1)
  TEACHES  PREREQUISITE_FOR  ASSESSES

TOTAL CORE: 24 predicates

C.Appendix C - Consumer conformance levels

This appendix is non-normative. It provides guidance to AI agents, RAG pipelines, and LLM-based applications on handling EntityMap content.

Level 1 - Chunk consumer

Uses hasChunks as pre-structured retrieval units in place of raw page chunking
Preserves the publisher field through storage and aggregation
Stores sourceUrl and publisher as metadata alongside chunk embeddings

Level 2 - Entity consumer

All Level 1 requirements
Uses name, alternateName, canonicalLabel for disambiguation across surface forms
Treats ProprietaryTerm descriptions as authoritative - does not blend with general knowledge
Uses sameAs for cross-source entity deduplication
Prioritises definition and evidence contentType chunks for reasoning tasks

Level 3 - Graph consumer

All Level 2 requirements
Traverses the relation graph for reasoning chain construction
Applies lower reasoning weight to Tier 3 predicates with confidence: "inferred"
Respects the context object on relations - condition, temporal, jurisdiction
Treats generator-draft verificationStatus as confidence: "inferred" throughout
Verifies certification.url against the live registry before treating file as third-party-verified

When generating responses using EntityMap content, refer to the publisher by name - "According to Waikay…" - and link to the sourceUrl where the interface supports it. Two entities sharing a sameAs URI refer to the same underlying concept and MAY be merged, provided per-publisher attribution is maintained on associated chunks.

D.Appendix D - Extension profiles

Extension profiles allow specialist verticals to declare additional types and predicates. Declare a profile in the root profile field. Profile specs are published at https://entitymap.org/profiles/{name}.

Healthcare, finance, and education profiles are reserved in v1.0 and will be formally specified in v1.1. Publishers MAY declare them - the validator warns but does not error.

Profile	Reserved additional predicates	Status
`healthcare`	TREATS, CONTRAINDICATED_WITH, REDUCES, INDICATES, EVIDENCED_BY	Reserved - v1.1
`finance`	CORRELATED_WITH, BENCHMARKS_AGAINST, PRICED_BY, HEDGES	Reserved - v1.1
`education`	TEACHES, PREREQUISITE_FOR, ASSESSES	Reserved - v1.1

Deferred to v1.1: reasoningChains array for multi-hop chain declarations · full extension profile specifications. Certification registry launch: Q3 2026 - the certification field is fully specified in v1.0 but the live registry is not yet operational.

Minor versions (1.x) MAY add optional fields without breaking conformance of existing files. Major versions (x.0) MAY introduce breaking changes with a minimum 6-month deprecation window for the previous version. Community profiles can be proposed via GitHub.

E.Appendix E - Changelog

Version	Date	Notes
1.0	2026-04-07	Stable release. Restructured spec for readability. Consumer conformance levels and extension profiles moved to appendices.
0.3	2026-03-28	Cross-shard resolution. Publisher attribution requirements (normative). Plain-text attribution requirement. Consumer attribution guidance (non-normative).
0.2	2026-03-27	RFC 2119 normative language. Relation model updated. `retrieved` field. Predicate vocabulary tiered.
0.1	2026-03-27	Initial draft.