The Visibility Code™— Unabridged

Author

David W. Bynon
Inventor of the WebMEM® Protocol
Founder, WebMEM

Publication Details

Version: 2.0
Originally Published: July 2025 as “Memory-First Publishing & Optimization”
Revised: August 2026
Original Publication Location: Prescott, AZ, USA

License

© 2025–2026 David W. Bynon. All rights reserved.

This work may be cited or excerpted for academic, technical, or strategic discussion with attribution. Implementation and commercial use of WebMEM may be subject to separate licensing terms.

WebMEM licensing and usage terms: WebMEM License

Abstract

This paper introduces a Knowledge Engineering for Answer Engines Framework (originally Memory-First Publishing and Optimization) for publishers, the body of work from which the WebMEM® Protocol emerged. Originally published in July 2025, the framework began with a central observation: as search and answer systems increasingly synthesize information directly, publishers can no longer assume that human-readable documents alone provide machines with sufficient structure to interpret, attribute, and apply published knowledge correctly.

The original framework explored this transition through the language of machine memory, retrieval conditioning, semantic reinforcement, and trust. Subsequent implementation revealed a more precise architectural problem. Publishers routinely possess identity, provenance, relationships, applicability, definitions, derivations, and other semantic structure that conventional web pages communicate indirectly to human readers but do not necessarily publish explicitly for machine interpretation.

The 2026 revision reframes Memory-First Publishing around that publisher-side responsibility.

Rather than attempting to prescribe or influence the internal memory, retrieval, ranking, or reasoning behavior of proprietary AI systems, the current framework asks a more durable question:

How should publishers represent the knowledge associated with a web resource so that machines can recover what the knowledge is about, where it came from, how its objects relate, and under what conditions it applies?

The resulting architecture separates human-facing presentation from a complementary machine-facing knowledge representation. It introduces the Semantic Data Template (SDT) as the page-level knowledge structure; Fragment Classes as semantic knowledge objects; explicit models for facts and assertions, identity, provenance, and relationships; a Resolution Model for determining applicable knowledge; Resolver architecture for publishing known resolution spaces; and serialization and conformance requirements for preserving those semantics in machine-readable form.

This revision therefore moves the framework from a theory of influencing machine retrieval behavior toward a protocol for publisher-controlled knowledge representation.

The objective is not to guarantee that a search engine, answer engine, language model, or agent will retrieve, cite, trust, rank, or remember published knowledge. Those behaviors remain under the control of consuming systems.

The objective is to ensure that when publishers make knowledge available to machines, they can publish more than isolated text and values. They can publish the semantic structure necessary to interpret that knowledge correctly.

Executive Summary

Digital publishing was built for a web in which machines primarily discovered and ranked documents and humans performed much of the remaining interpretation.

That division of labor is changing.

Search engines, answer engines, language models, and agentic systems increasingly identify entities, retrieve facts, combine information from multiple sources, compare alternatives, and synthesize answers before a user necessarily visits the originating document.

This creates a publishing problem that traditional search optimization and conventional structured markup do not fully address.

A publisher may possess the correct data and still leave machines to reconstruct critical meaning from prose, tables, URL patterns, navigation, application logic, document proximity, or other presentation structures designed primarily for humans.

The missing information may include:

• what entity an assertion actually describes;
• which source supports the assertion;
• whether a value was source-observed or publisher-derived;
• how two identifiable objects relate;
• which geographic, temporal, jurisdictional, or other context governs applicability;
• which objects belong to an applicable collection or option space;
• and which canonical resource allows resolution to continue.

These are not primarily ranking problems.

They are knowledge representation and resolution problems.

Memory-First Publishing began in 2025 as an attempt to understand how publishers might structure information for an emerging environment of AI-mediated retrieval and synthesis. The original work introduced Semantic Digests, Semantic Data Templates, semantic bindings, provenance structures, defined terminology, cross-surface reinforcement, retrieval feedback mechanisms, and other experimental concepts.

Production implementation during the following year clarified which ideas belonged to a durable publisher-side architecture and which depended too heavily on assumptions about proprietary machine behavior.

The resulting 2026 framework establishes a clearer boundary:

Publisher
    ↓
Human-Facing Representation
+
Machine-Facing Knowledge Representation
    ↓
WebMEM Boundary
    ↓
Consuming System
    ↓
Retrieval / Ranking / Reasoning / Synthesis / Citation

WebMEM governs the publisher side of that boundary.

It does not prescribe the internal behavior of the consuming system.

Key components of the current framework include:

• Semantic Data Templates (SDTs): Page-level machine-facing representations of the knowledge associated with a web resource.
• Fragment Classes: Modular knowledge objects that declare the semantic role of the knowledge they contain, including factual data, derived statistics, identity, indexes, definitions, policies, procedures, directories, explanations, and other structures.
• Facts and Assertions: Explicit publisher statements about identified subjects within defined semantic context.
• Identity: The structures required to preserve what entity, fragment, dataset, or resource the represented knowledge actually describes.
• Provenance: Recoverable lineage connecting published knowledge to its sources, transformations, derivations, and responsible actors.
• Relationships: Explicit semantic connections among identifiable objects rather than relationships inferred from document proximity, navigation, or naming.
• Resolution: The process through which identity, relationships, context, and applicability converge on the knowledge object, entity, option space, or resource appropriate to an information need.
• Resolvers: Publisher-defined information resources that expose known resolution structure around meaningful identifiers, entities, collections, or contexts.
• Serialization: The concrete representation of the WebMEM knowledge model, currently implemented through HTML-in-HTML using inert <template> containers and semantic HTML bindings.
• Conformance: Requirements ensuring that WebMEM representations preserve applicable semantics rather than merely reproducing WebMEM-like syntax.

The framework applies broadly to domains in which identity, evidence, applicability, and factual precision matter, including healthcare, government, law, finance, education, product information, and other structured information systems.

The central proposition is no longer that publishers should attempt to condition machine memory.

It is simpler and more defensible:

Publishers should explicitly publish the semantic structure they already know.

If a publisher knows which entity a fact describes, that identity should not need to be reconstructed from prose.

If the publisher knows where a fact came from, its provenance should travel with the knowledge.

If the publisher knows how two entities relate, the relationship should not need to be inferred from links or document proximity.

If the publisher knows that a value applies only to a particular geography, period, segment, jurisdiction, or population, that applicability should survive machine extraction.

If the publisher knows the structure required to move from an identifier or context to an applicable information space, that resolution structure can itself be published.

This reframes the emerging discipline from retrieval engineering toward publisher-side knowledge engineering for answer engines.

The publisher does not control what a machine remembers.

The publisher does control what it publishes for the machine to interpret.

This marks the evolution:

• From documents → to knowledge representation
• From markup → to semantics
• From retrieval → to resolution
• From inferred context → to explicit structure
• From machine-conditioning claims → to publisher-side responsibility

The machine’s job changed. The publisher’s job must change with it.