RESEARCHED · 2026.09.12-g40 · not deployed
Graph systems — research and implementation expansion
READ THIS FIRST
This expansion defines 40 selected graph types, representations, and practical graph views. It is not a claim that there are exactly 40 relevant graphs in computer science, that search engines disclose all of their graphs, or that these 40 views are 40 independent ranking systems. Many are different views over one source-linked data model. The names and priorities organize this library; they are not professional certifications or provider-mandated departments.
There are four different kinds of authority here:
- Documented provider systems, such as Google's Knowledge Graph and Shopping Graph.
- Standards and protocols, such as RDF, JSON-LD, PROV-O, SKOS, SHACL, and Open Graph.
- Published methods, such as GraphRAG and HNSW approximate-neighbor indexing.
- This library's proposed analytical and operating views, including an E-E-A-T evidence graph, a SERP observation graph, and a release-evidence graph.
A standard can define valid structure without certifying truth. A provider can describe a system without exposing its current internals. A useful internal graph can improve consistency without being a ranking factor. Sources S01–S49 are listed at the end and in Research/sources.json; individual cards identify their relevant source basis. Standards are pinned references, not claims that no later specification exists.
1. WHAT THE SUPPLIED LIBRARY ALREADY HAS
The supplied TXT is a conversion of 135 embedded documents from the HTML release 2026.09.09-d26. Its R11 charter already requires stable identifiers, truthful Person/business/document distinctions, exact identity links, graph consistency, and claim provenance. The public entity plan already prohibits inventing organizations, credentials, or Wikidata identities. Those foundations are retained.
The supplied master index lists framework-eeat.md, SERP-Optimization.md, and framework-entitysalience.md as legacy, not fully verified. Those index entries are not equivalent to complete, freshly researched graph lessons. The export also references companion artifacts that are not included in the TXT. This expansion does not claim to have recovered or revalidated an unseen original archive.
The additions here are a typed graph catalog; practical personal/business modeling; explicit E-E-A-T and SERP evidence views; private AI-retrieval and governance views; examples and blocked templates; and executable offline package checks. No commercial tier, role number, or original text has been silently rewritten. R11 remains the graph/structured-data lead. R14 governs factual claims; R08 URL contracts; R12 AI observations; R16 measurement; R20 privacy; R24 integration; R25 authorized release; R26 scoped acceptance.
2. START WITH A RELATIONSHIP, NOT A PICTURE
A graph contains nodes and relationships. A typed semantic example is:
[article] --author--> [person] [article] --publisher--> [organization] [service] --provider--> [organization]
In RDF, the basic unit is a subject–predicate–object triple; datasets can separate graphs. JSON-LD is one way to serialize linked data using JSON. A visual network diagram is only a view: the useful asset is the inspectable identity, relationship, evidence, and version record. [S12, S13]
For this library, a relationship that makes a material real-world claim must also identify its basis. "Joseph founded Organization A" is not merely a line to draw. It is an assertion requiring the right person, the right organization, an actual founding relationship, and evidence. Ownership, founding, employment, operating a website, publishing an article, and using a brand must not be substituted for one another.
Use a durable entity registry and a separate claim registry. The registry distinguishes things from their documents. A person may have multiple profile pages without becoming multiple people. An organization may publish several websites without those sites becoming the same page. A brand may identify a commercial offering without being a separate legal organization.
Do not infer that an absent fact is false. An unknown founder, office address, credential, or legal name remains unknown. A graph that leaves unsupported attributes empty is preferable to one that invents a complete-looking biography.
3. PERSONAL AND BUSINESS MODEL FOR JOSEPH
Joseph's name and the domains ThatDevPro.com and ThatDeveloperGuy.com are user-supplied context. This task has not independently audited their live pages, legal structures, official profiles, ownership evidence, or search-engine recognition. Do not turn the following decision process into published claims without verification.
First establish Joseph's approved public identity and its canonical profile home. Then inventory each actual operating organization, trading name, brand, website, and service. Decide which distinctions are real before choosing types. Keep a register of ambiguities instead of guessing that the two domains are either the same legal business or separate legal businesses.
The proposed logical model is:
Joseph (Person)
↔ actual authored work and approved profile pages
↔ only verified professional or business relationships
Operating organization(s), once established
→ real brand(s), where applicable
→ actual service(s)
→ website(s) and published work
ThatDevPro.com / ThatDeveloperGuy.com (websites)
→ their verified publisher/operator
→ distinct pages and purpose
Choose one stable public identity for a person or organization and reuse it consistently wherever that exact entity is represented. A page's canonical URL and the identity of its subject are separate decisions. Do not migrate an existing identity or page solely because this document shows a cleaner-looking example.
Schema.org's sameAs identifies the same item through an unambiguous reference. It is not an "all my websites" field, an ownership relationship, or a way to borrow authority from related Wikipedia topics. [S04] The packet's example deliberately omits sameAs because no actual third-party identity URLs have been verified here.
For an appropriate biography page, ProfilePage can point through mainEntity to the Person. A real article can reference its actual author and publisher. A real service can reference its provider. ProfilePage is not a universal type for every business landing page. [S03, S07]
Implementation sequence proposed for Joseph:
- Approve actual person, organization, brand, website, and service identities.
- Record evidence and the canonical-home decision before minting production IDs.
- Render approved facts consistently in visible HTML and applicable JSON-LD.
- Check existing public profiles for exact identity, then add only justified identity references.
- Test all selected representations and record engine observations separately.
The general library should retain this decision process and a neutral synthetic example, not hard-code Joseph's private facts into every future client build.
4. THE E-E-A-T GRAPH THAT IS USEFUL TO BUILD
Google discusses experience, expertise, authoritativeness, and trustworthiness as qualities associated with useful content; E-E-A-T itself is not a specific ranking factor. Google does not supply an "E-E-A-T graph" specification or public score for this library to implement. [S01]
Our proposed E-E-A-T evidence view instead asks a concrete question: what supports this particular experience or expertise claim?
claim about capability
→ actual work performed
→ original evidence with scope and date
→ an attributable explanation
→ authorized publication
→ independent corroboration, when it genuinely exists
Examples of defensible inputs are a documented repair, an original design process, a reproducible test, a relevant qualification actually held, or a correction made when evidence changed. A case study must distinguish observed outcomes from causality. A public methodology article can teach a process without pretending that a client engagement occurred.
Store the source owner and whether each mention is owned, paid, syndicated, or independently earned. Two sites owned by the same person can truthfully describe that person's work, but they should not be counted as two independent endorsements merely because they use different domains. The anti-circularity rule is part of this library's evidence design.
Use the view to find unsupported assertions, stale biographies, overbroad claims, missing author responsibility, and public evidence gaps. Do not convert it into an invented Google score or require artificial citations for every sentence. The objective is justified claims and useful instruction, not graph density.
5. THE SERP GRAPH THAT CAN ACTUALLY BE MEASURED
SERP means search engine results page. The proposed graph is an observation model, not an exposed internal search-engine graph. Each result appearance belongs to a snapshot; the same URL appearing twice or in different features should not be flattened into a timeless "ranks for" edge.
Record engine, surface, query, observation time with time zone, language, location setting at the least sensitive useful resolution, device class, collection method, personalization assumptions, result feature, position definition, raw URL, and evidence reference. Unknown values remain null. Do not infer the user's precise location or private search history.
Represent the relationship as:
snapshot --for-query--> query snapshot --contains--> result occurrence result occurrence --points-to--> URL result occurrence --has-feature--> organic / local / ad / other observed feature
Query-to-document graphs are naturally modeled with two node sets; the detailed snapshot design here is our own use of that representation. [S40]
To compare two observed result sets A and B, a simple descriptive overlap is:
Jaccard(A, B) = |A intersection B| / |A union B|
Define whether the members are full URLs, canonical pages, or domains before calculating. Do not mix definitions between comparisons. If the union is empty, record the result as undefined, not as perfect similarity or zero visibility. A position-sensitive metric needs its own explicitly declared weighting and is still not an engine's proprietary weight.
Use overlap to discover candidate clusters and competing answers, then inspect user intent and content. A competitor sharing several results does not prove its schema, page length, or internal-link pattern caused the ranking. A sampled absence is not proof that a website never appears anywhere.
6. AI CITATIONS NEED THEIR OWN OBSERVATION LAYER
Bing's current AI Performance documentation supports analysis of grounding-query and cited-page relationships and explicitly separates citation observations from ranking, authority, and traffic. Grounding phrases must not be relabeled as the original user prompts. Coverage, aggregation, and denominators matter. [S22, S23]
Keep separate record types for an actual captured answer, an aggregated report row, a citation occurrence, and a source-support assessment. An aggregated report must not be converted into fabricated individual answers. A page cited by an answer may support only one part of that answer; inspect the relevant source passage before asserting support.
Google's current guidance also points to a Generative AI performance report in Search Console. Reconfirm the available report, account coverage, and field definitions before building an importer; a private graph cannot substitute for actual account access. [S21]
Recommended outputs are a dated citation-observation table, a source/claim support map, and a separate referral/lead view. Keep unavailable evidence explicit. Never multiply citation counts by assumed click-through or conversion rates and present the product as measured customers.
OpenAI documents different access roles for search, training, and user-directed retrieval. Treat those as distinct controls and observations. A search crawler's permitted access does not prove that a page was selected or that a private library was learned. [S27]
7. WHAT PUBLIC GRAPHS AND APIS DO NOT GIVE YOU
Google's Knowledge Graph is a documented entity system. Its older Knowledge Graph Search API is read-only, returns matching entities rather than a connected-graph export, includes a migration notice, and warns against a production-critical dependency. An API response score is not a website authority or E-E-A-T measurement. Recheck the current replacement product and terms before designing a new integration. [S10, S11]
Google has also described Shopping Graph and, historically, a Topic Layer associated with Knowledge Graph. These are useful concepts to teach, but historical announcements are not complete current architecture specifications. Merchant data and relevant public content are inputs through supported mechanisms, not an interface for writing arbitrary ranking relationships. [S31, S44]
Wikidata and OpenAlex offer useful public entity and scholarly-data resources. Use them for reference discovery and identifier reconciliation with source and identity checks. Do not create misleading entries, copy unsupported attributes, or treat inclusion as an endorsement. [S36, S37]
Open Graph is the social-preview protocol, not Google Knowledge Graph. Its metadata should be accurate and useful to sharing consumers, but it does not replace your entity or evidence model. [S35]
8. PUBLIC PUBLISHING VERSUS PRIVATE AI RETRIEVAL
Maintain two deliberate projections of the approved data:
PUBLIC Actual useful pages, approved identity facts, accurate bylines, applicable structured data, images, navigation, and supported feeds when relevant.
PRIVATE Raw evidence, client records, ownership documentation, review decisions, graph-extraction candidates, source lineage, permissions, tests, and workflows.
Do not publish the private corpus, AGENTS instructions, logs, access tokens, client messages, or inferred relationships to make a graph easier for crawlers to discover. A robots directive is not a substitute for access control.
Google's generative-search guidance does not require special AI files or special schema. Private GraphRAG chunking is an internal retrieval choice, not a reason to break every public article into tiny pages. A custom public /graph.json endpoint should exist only for a named consumer and a reviewed contract, not because it supposedly forces all engines to use it. [S21]
The practical architecture is one governed registry with multiple adapters. Generate selected public facts into HTML and suitable structured data. Use private source-backed records for AI retrieval and audits. Keep public and private storage and authorization boundaries enforceable, not merely decorative visibility labels.
9. GRAPH-RAG IS AN OPTIONAL RETRIEVAL METHOD, NOT AN SEO SWITCH
Microsoft GraphRAG organizes source material using extracted entities and relationships and can combine graph-linked information with original text units. Its documentation describes multiple retrieval strategies. [S24, S25]
For this library, retain the source text and version first. Extract candidate entities, resolve them carefully, and store candidate relationships with provenance. Generated summaries are derived artifacts and inherit restrictions from their sources. They must not become new independent evidence just because another model wrote them.
Evaluate a pilot against ordinary text search and a vector-retrieval baseline using tasks such as:
- Find the current claim and supporting source for one structured-data rule.
- Distinguish a person from an organization with a similar name.
- Retrieve an older rule only when the question concerns the older date.
- Refuse a request to expose private evidence in public markup.
- Identify conflicting source statements without inventing a resolution.
Measure answer support, entity-resolution errors, missed applicable sources, restricted-data leakage, latency, and cost. Do not assume the graph approach wins every task. An HNSW index is a graph of approximate vector-neighbor relationships, not the same thing as a graph of verified facts. [S26]
10. MACHINE-READABLE CLAIM CONTRACT
The supplied claim.schema.json is a library record schema, not Schema.org markup. It keeps candidate facts separate from publication approval. The object is either a referenced entity or a literal; the two forms are not interchangeable. Example and candidate records remain intentionally blocked for publication-readiness purposes.
Required concepts include:
claim ID and namespace-qualified predicate; subject and typed object; evidence class and source locators; observation time and effective interval, when known; status: candidate, supported, disputed, retired, or unknown; confidence rationale, not an uncalibrated ranking-like number; owner/source independence classification; verification state and reviewer record; private/public-approval state and actual authority to publish; example flag, review date, and dependent graph views.
The schema can test form. The included Python helper tests a deliberately limited structural readiness policy. Neither can verify that a reviewer is real, that evidence supports a statement, or that a client authorized publication. Human or otherwise genuinely authorized review must evaluate actual evidence. The helper always reports authorization_granted=false.
Changing a candidate to "supported" by editing JSON does not create support. A source URL without a relevant locator does not establish a claim. A copied article with a different URL is not independent corroboration. A missing field must not be filled with a plausible guess solely to satisfy validation.
11. ACCEPTANCE LAYERS
Use separate outcomes for separate checks:
A. Record validity: JSON parses; required fields and types are present. B. Identity consistency: IDs are unique and refer to intended distinct entities. C. Relationship validity: selected predicates fit the intended meaning. D. Evidence review: material assertions have relevant source-backed support. E. Temporal applicability: historical or superseded instructions are not used as current. F. Privacy and permission: the actual system prevents unauthorized disclosure. G. Page agreement: public structured data agrees with visible approved facts. H. Feature eligibility: the current targeted provider requirements are satisfied. I. Observed discovery: actual crawl/index/result/citation evidence is recorded. J. Business outcome: actual qualified inquiries or revenue are measured separately.
SHACL provides a standards-based way to check graph constraints; this package supplies a JSON Schema rather than claiming to ship a full SHACL implementation. Neither syntax nor shape conformance proves factual truth. [S16] Public feature validation is a separate task. [S08]
For each required check, retain PASS, FAIL, BLOCKED, or NOT_APPLICABLE with rationale. Missing live access is BLOCKED, not a fabricated pass. A local fixture is not production evidence, and a second pass by the same assistant is not an independent audit.
12. EVALUATION WITHOUT FAKE AUTHORITY SCORES
These are proposed internal quality measures, not search-engine metrics:
Provenance coverage: supported material public claims divided by material public claims in the declared sample. Report the sample, exclusions, and manual-review method.
Identity-resolution precision: reviewed correct accepted matches divided by reviewed accepted matches. Keep ambiguous and unreviewed matches outside the numerator and report their counts separately.
Reference closure: internal references resolving to expected records divided by tested internal references. External URLs require separate checks; they are not unresolved just because they are not nodes in the local graph.
Staleness exposure: number of current-use artifacts depending on claims whose review is overdue under the chosen internal policy. Overdue means review needed, not automatically false.
Contradiction burden: unresolved material disputes and their dependent outputs. Do not hide them in one average confidence score.
Retrieval support rate: sampled answers whose material assertions are supported by the retrieved sources, using an explicit review rubric. This needs actual evaluated answers, not a theoretical score.
For a zero denominator, report undefined or not applicable with reason rather than manufacturing a perfect percentage. Metrics should identify useful next actions, not become optimization targets detached from truth.
13. CURRICULUM AND WORKED CHECKS
Lesson 1 — Identity versus document
Exercise: one developer has two websites and three social profiles. How many people exist? Expected answer: the records may describe one person, but exact matches require evidence. Websites and profile pages are separate entities/documents. Do not infer the number of legal businesses from the number of domains.
Lesson 2 — Exact identity versus relationship
Exercise: two brands share an owner. Should they be linked with sameAs? Expected answer: no, not merely for shared ownership. Record the actual relationship and keep their identities distinct. Use sameAs only when the reference unambiguously identifies the same item. [S04]
Lesson 3 — Provenance versus repetition
Exercise: six articles repeat one press release. How many independent sources support its central claim? Expected answer: this dataset shows one underlying origin unless independent investigation is demonstrated. Six publication URLs alone do not establish six independent confirmations.
Lesson 4 — Evidence versus E-E-A-T scoring
Exercise: an AI writes "internationally recognized expert" because the entity graph has many edges. Expected answer: reject or rewrite the claim unless real evidence supports it. A graph's density is not an expertise credential or an official E-E-A-T score.
Lesson 5 — Snapshot versus timeless ranking
Exercise: a service page is absent from one mobile result snapshot in one city. Expected answer: report that bounded observation. It does not establish absence from every search or a causal reason for the result.
Lesson 6 — Citation versus business outcome
Exercise: a report shows 100 citations but no referral or CRM data. Expected answer: report the citations with scope. Referral sessions, leads, and revenue are unavailable, not zero and not estimated actuals.
Lesson 7 — Shape versus truth
Exercise: a valid JSON-LD graph says an invented person earned an invented award. Expected answer: syntax may pass, but evidence/publication checks fail. A shape validator cannot establish the real-world claim.
Lesson 8 — Retrieval versus publication
Exercise: GraphRAG finds a useful client email proving an outcome. Expected answer: use only within authorized private scope; do not publish or disclose it without actual permission and redaction review. A derived summary can still leak confidential content.
Lesson 9 — Time and authority
Exercise: a fresh blog contradicts an older applicable official specification. Expected answer: inspect scope, effective dates, and source authority. Freshness alone does not resolve the conflict.
Lesson 10 — Delivery versus discovery
Exercise: public JSON-LD renders correctly, but there is no engine observation. Expected answer: report successful delivery checks only. Indexing, panel inclusion, citations, and rankings remain unobserved.
14. IMPLEMENTATION ORDER
Priority P0: personal identity; actual organizations/brands; cross-domain relationships; entity resolution; authorship; claim provenance; internal navigation; canonical policy; crawl/render evidence; public representation consistency; acceptance; permissions. Deliverable: a small approved registry, source-backed claims, limited public projection, and an inspectable test report. Do not publish candidate identities merely to make the graph look complete.
Priority P1: E-E-A-T evidence, reputation, topics, query/page relevance, SERP and AI-citation observation, conversion semantics, Open Graph, curriculum, and workflow routing. Local and migration modules activate when applicable. Deliverable: useful explanations and evidence, working observation definitions, and isolated public/private data flows. A module marked applicable still needs actual data and tests.
Priority P2: private GraphRAG pilot, vector-neighbor indexing when needed, temporal/supersession controls, contradictions, source-impact mapping, citation projections, and co-occurrence analysis. Deliverable: measured retrieval comparisons and a maintenance process. Do not choose an infrastructure stack by fashion or a hypothetical SEO boost.
Priority P3: commerce, localization, media, professional/social graphs, scholarly data, and events as the actual website or teaching objective requires. Deliverable: specialized contracts and truthful data, not empty graph templates represented as deployed capabilities.
These priorities are a proposed dependency order, not provider mandates or a guarantee of search improvement. Preserve original 26-role routing rather than launching 40 independent agents or creating 40 databases.
15. MAINTENANCE AND SOURCE GOVERNANCE
Every source record needs a title, publisher, exact URL, reviewed date, scope, and limitations. The supplied source register does not pretend that a review date is a publication date or that source pages were archived. Add actual effective dates when established.
Proposed maintenance policy: re-open fast-changing platform guidance before a deployment decision that depends on it; run a periodic change review for provider features; review identity claims when the owner reports a change; and review evidence permissions before every new public use. Select actual scheduling and ownership in the host project. Nothing in this packet installs a scheduler or background monitor.
On change, record the specific altered claim, evaluate applicability, mark affected artifacts for review, test proposed revisions, and publish a bounded changelog. A changed webpage hash is only a signal to inspect; navigation edits are not necessarily policy changes. Keep superseded lessons available as history with their dates, but out of default current-use retrieval.
16. AI PROJECT INTEGRATION CONTRACT
Use the existing root AGENTS.md and role router. Add a reference to this expansion only after merging with project instructions; do not replace a real AGENTS.md automatically. This packet includes an integration snippet, not a remote configuration change.
An AI task should receive: the authorized objective; verified site facts; relevant graph IDs; selected role owners; actual read/write limits; current sources; desired public/private outputs; acceptance criteria; and rollback requirements for any approved change. Retrieve only relevant lessons and records rather than loading the whole library as unquestioned authority.
The assistant must distinguish proposed work, actual code changes, locally executed checks, untested external behavior, and live engine observations. It may not fill missing evidence with fictional outputs, self-granted approval, fabricated credentials, or a claim that changing persona made its review independent.
17. PACKAGE SCOPE AND LIMITATIONS
This release adds the curriculum and machine-readable starting structures. It does not build a production graph database, install GraphRAG, access Joseph's accounts, verify all public profiles, submit search data, deploy markup, edit a live site, or recertify the original 135 documents. The 40 cards are source-linked implementation guides, not completed deployments.
The offline tests inspect this package's records, examples, and limited readiness rules. They do not run live Rich Results tests, search-engine indexing tests, real browser journeys, private-source permissions, or an external independent audit. Those remain requirements for a real implementation.
18. NAMED RESEARCH SYSTEMS AND ALGORITHMS TO KNOW
These names belong in the library because they explain several real graph-based approaches that marketing language often blends together. They are a methods appendix, not five new disclosed ranking factors. Map them into the appropriate Gxx views rather than adding an agent, endpoint, or public score for every name.
Knowledge Vault — probabilistic knowledge fusion. The 2014 Google Research paper combines extracted web information with prior knowledge and models uncertainty in the fused facts. This is a useful historical basis for teaching extraction, reconciliation, confidence calibration, and source fusion. It is not the same claim as "my site has been entered in Knowledge Vault." [S46] Library use: connect G04 entity resolution, G09 document/entity mappings, G10 provenance, and G30 contradictions. Keep uncertain extractions out of approved public facts. Calibrated probabilities in a research model are different from an assistant inventing a confidence percentage.
Knowledge-Based Trust (KBT) — fact-based source trust research. The 2015 paper investigates estimating source trustworthiness from extracted facts while separating extraction mistakes from errors in the source itself. [S45] Library use: teach why evidence quality and the reliability of extraction must both be examined. A bad parser can make a correct source look inconsistent; a frequently repeated assertion can still be wrong. KBT is not interchangeable with E-E-A-T, link-based trust, or a consumer-visible website score. Do not claim a modern engine uses this exact paper without separate deployment evidence.
TrustRank — seed-guided link analysis for spam detection. The 2004 paper uses a small expert-reviewed seed set and the web's link structure to identify other pages likely to be reputable. The reviewed paper is by researchers affiliated with Stanford and Yahoo, and its experiments are historical. [S49] Library use: attach this method to G14 external link analysis and an explicitly labeled research lesson on trust propagation. Explain seed dependence and probabilistic assumptions. Do not substitute "all government sites" or "all university sites" for a demonstrated seed set, and do not translate proximity into a guaranteed modern ranking benefit. The paper is not evidence that the original TrustRank algorithm currently governs Google Search.
HITS — hubs and authorities. HITS distinguishes hub scores from authority scores for a supplied linked graph; the NetworkX documentation exposes an implementation. [S47] Library use: compare it with PageRank in a controlled G13/G14 lesson. Show that two algorithms applied to the same small graph answer different mathematical questions. Neither supplies Google's hidden scores. A node labeled "authority" by an algorithm has not acquired a real-world credential or permission to make unsupported claims.
Query–entity click graphs. Microsoft Research's 2011 Jigs and Lures work models associations between queries and typed entities using click-graph information. [S48] Library use: enrich G21 query relevance and G26 measurement teaching with a distinction between a query-to-result appearance and an observed click association. Only use actually available, authorized data. Do not invent click logs from rankings or infer a click from a displayed citation. Historical experimental gains are not a promised improvement on another business's site.
A compact comparison:
Knowledge graph: entities and semantic relationships. Web-link graph: pages and hyperlinks. Click graph: observed query/item or user/item interactions. Trust-propagation model: assumptions and labels propagated across a network. Probabilistic knowledge fusion: combining uncertain extracted assertions. E-E-A-T evidence view: this library's map from claims to real support. SERP observation view: this library's dated record of displayed results. Vector-neighbor graph: proximity under an embedding representation.
Do not transfer a score across these categories. A high local hub score, an extracted-fact probability, a vector similarity, a citation count, and an editorial claim of expertise are not measurements of the same thing. Every algorithm lesson must identify its graph definition, data scope, parameters, assumptions, evaluation set, and deployment status.
Graph catalogue index
- [G01 — Personal identity and profile graph](/library/graph-systems/g01-personal-identity-and-profile-graph/) — P0; standards-backed library view
- [G02 — Organization, business, and brand graph](/library/graph-systems/g02-organization-business-and-brand-graph/) — P0; standards-backed library view
- [G03 — Cross-domain operating and ownership graph](/library/graph-systems/g03-cross-domain-operating-and-ownership-graph/) — P0; library analytical view
- [G04 — Entity resolution and reconciliation graph](/library/graph-systems/g04-entity-resolution-and-reconciliation-graph/) — P0; library analytical view
- [G05 — Local business, place, and service-area graph](/library/graph-systems/g05-local-business-place-and-service-area-graph/) — P1_IF_APPLICABLE; standards-backed library view
- [G06 — Product, offer, merchant, and inventory graph](/library/graph-systems/g06-product-offer-merchant-and-inventory-graph/) — P3_IF_APPLICABLE; standards-backed library view
- [G07 — Authorship and editorial contribution graph](/library/graph-systems/g07-authorship-and-editorial-contribution-graph/) — P0; standards-backed library view
- [G08 — Topic, ontology, and taxonomy graph](/library/graph-systems/g08-topic-ontology-and-taxonomy-graph/) — P1; standards-backed library view
- [G09 — Document, passage, and entity graph](/library/graph-systems/g09-document-passage-and-entity-graph/) — P1; library analytical view
- [G10 — Claim, evidence, and provenance graph](/library/graph-systems/g10-claim-evidence-and-provenance-graph/) — P0; standards-backed library view
- [G11 — E-E-A-T evidence graph](/library/graph-systems/g11-e-e-a-t-evidence-graph/) — P1; custom library evidence view; not a Google graph specification
- [G12 — Reputation, review, and corroboration graph](/library/graph-systems/g12-reputation-review-and-corroboration-graph/) — P1; library analytical view
- [G13 — Internal link and navigation graph](/library/graph-systems/g13-internal-link-and-navigation-graph/) — P0; library analytical view of documented web links
- [G14 — External web-link graph](/library/graph-systems/g14-external-web-link-graph/) — P1; documented link-analysis concept; local reconstruction
- [G15 — Citation, co-citation, and bibliographic-coupling graph](/library/graph-systems/g15-citation-co-citation-and-bibliographic-coupling-graph/) — P2; research/analysis view
- [G16 — Canonical and duplicate-document graph](/library/graph-systems/g16-canonical-and-duplicate-document-graph/) — P0; library view of documented URL policy
- [G17 — Redirect and migration graph](/library/graph-systems/g17-redirect-and-migration-graph/) — P1_IF_APPLICABLE; library engineering view
- [G18 — Hreflang and locale-alternative graph](/library/graph-systems/g18-hreflang-and-locale-alternative-graph/) — P3_IF_APPLICABLE; library view of documented localization signals
- [G19 — Crawl, rendering, and resource-dependency graph](/library/graph-systems/g19-crawl-rendering-and-resource-dependency-graph/) — P0; library engineering view
- [G20 — Structured-data consistency graph](/library/graph-systems/g20-structured-data-consistency-graph/) — P0; serialization/validation view
- [G21 — Query, intent, and page-relevance graph](/library/graph-systems/g21-query-intent-and-page-relevance-graph/) — P1; library analytical view
- [G22 — SERP observation and competitor-overlap graph](/library/graph-systems/g22-serp-observation-and-competitor-overlap-graph/) — P1; custom observational view; not an official SERP graph
- [G23 — Question, task, and query-fan-out map](/library/graph-systems/g23-question-task-and-query-fan-out-map/) — P1; custom planning/observation view
- [G24 — Entity-mention and co-occurrence graph](/library/graph-systems/g24-entity-mention-and-co-occurrence-graph/) — P2; library analytical view
- [G25 — AI answer, grounding, and citation graph](/library/graph-systems/g25-ai-answer-grounding-and-citation-graph/) — P1; custom observation view of documented report data
- [G26 — Search-to-lead and customer-journey graph](/library/graph-systems/g26-search-to-lead-and-customer-journey-graph/) — P1; custom measurement view
- [G27 — GraphRAG and owned-corpus retrieval graph](/library/graph-systems/g27-graphrag-and-owned-corpus-retrieval-graph/) — P2; documented method with library-specific adaptation
- [G28 — Embedding-similarity and nearest-neighbor graph](/library/graph-systems/g28-embedding-similarity-and-nearest-neighbor-graph/) — P2; published algorithmic method
- [G29 — Temporal, version, and supersession graph](/library/graph-systems/g29-temporal-version-and-supersession-graph/) — P2; library governance view
- [G30 — Contradiction, uncertainty, and dispute graph](/library/graph-systems/g30-contradiction-uncertainty-and-dispute-graph/) — P2; library governance view
- [G31 — Source-dependency and change-impact graph](/library/graph-systems/g31-source-dependency-and-change-impact-graph/) — P2; library governance view
- [G32 — Instruction, capability, and workflow dependency graph](/library/graph-systems/g32-instruction-capability-and-workflow-dependency-graph/) — P1; private library operating design
- [G33 — Requirement, test, evidence, and acceptance graph](/library/graph-systems/g33-requirement-test-evidence-and-acceptance-graph/) — P0; private library operating design
- [G34 — Permissions, consent, and disclosure graph](/library/graph-systems/g34-permissions-consent-and-disclosure-graph/) — P0; policy-description and private governance view
- [G35 — Multimedia, transcript, and rights graph](/library/graph-systems/g35-multimedia-transcript-and-rights-graph/) — P3_IF_APPLICABLE; standards-informed library view
- [G36 — Open Graph social-preview metadata](/library/graph-systems/g36-open-graph-social-preview-metadata/) — P1; documented protocol; naming distinction
- [G37 — Social, professional, and interaction graphs](/library/graph-systems/g37-social-professional-and-interaction-graphs/) — P3_IF_APPLICABLE; documented graph family; library analysis must be scoped
- [G38 — Scholarly and research graph](/library/graph-systems/g38-scholarly-and-research-graph/) — P3_IF_APPLICABLE; documented public-data graph family
- [G39 — Event, venue, organizer, and time graph](/library/graph-systems/g39-event-venue-organizer-and-time-graph/) — P3_IF_APPLICABLE; standards-backed library view
- [G40 — Learning prerequisite and curriculum graph](/library/graph-systems/g40-learning-prerequisite-and-curriculum-graph/) — P1; custom teaching design
Local validation summary
28 standard-library unit tests passed. Package integrity checks passed. The record schema and blocked template validated, and four malformed/publication-conflicting variants were rejected using jsonschema 4.26.0. These results describe local artifacts only. No website, public profile, search account, rich-result service, or independent external audit was tested.