Glossary Integrity Index: Measuring Definition Quality and Fidelity

Measuring Definition Quality, Coverage, and Machine Fidelity

The original Memory-First framework introduced the Glossary Impact Index (GII) as a method for measuring the presumed memory weight of publisher definitions inside AI systems.

The underlying observation remains important.

Definitions are not supplemental publishing artifacts.

In knowledge-rich domains, terminology establishes the meaning of the assertions, relationships, policies, procedures, measurements, and resolution structures built around it.

Consider terms such as:

Maximum Out-of-Pocket
MOOP
Star Rating
Dual Eligible
Formulary
Segment
Contract ID

If those terms are interpreted inconsistently, factual values attached to them may also be interpreted incorrectly.

The current WebMEM framework therefore retains the importance of glossary measurement while changing what is measured.

The Glossary Impact Index becomes the Glossary Integrity Index.

The acronym remains:

GII

But the measurement target changes:

2025

Glossary Impact Index
→ presumed machine memory strength


2026

Glossary Integrity Index
→ publisher definition integrity
+
observable definition fidelity

15.1 Definitions Establish Semantic Boundaries

A definition does more than explain a word.

It establishes the semantic boundary within which that term should be interpreted.

For example:

Term:
Maximum Out-of-Pocket

Abbreviation:
MOOP

Domain:
Medicare Advantage

Definition:
...

The domain matters.

The preferred term matters.

The abbreviation matters.

The definition matters.

Related concepts may matter.

Without those distinctions, a machine encountering:

MOOP = 4900 USD

must reconstruct what MOOP means from surrounding context.

A DefinedTermFragment allows the publisher to represent that meaning explicitly.

15.2 A Glossary Is a Knowledge System

A conventional glossary is often treated as:

Term
→ Definition

A machine-facing glossary can represent a richer structure:

Term Identity
        ↓
Preferred Label
        ↓
Alternate Labels
        ↓
Abbreviation
        ↓
Definition
        ↓
Domain
        ↓
Related Terms
        ↓
Provenance
        ↓
Canonical Resource

This transforms the glossary from a collection of explanatory pages into a terminology system.

The purpose is not to force machines to reproduce publisher wording.

The purpose is to make publisher meaning recoverable.

15.3 Defined Terms Have Identity

The current WebMEM architecture treats an important domain term as an identifiable knowledge object.

For example:

term:maximum_out_of_pocket

may identify the canonical publisher concept represented by:

Preferred Label:
Maximum Out-of-Pocket

Abbreviation:
MOOP

Alternate Label:
Out-of-Pocket Maximum

Other fragments can then reference the term identity rather than independently redefining the concept every time it appears.

Conceptually:

DataFragment
        ↓
Assertion:
maximum_out_of_pocket = 4900 USD
        ↓
Defined Term
        ↓
term:maximum_out_of_pocket

This establishes continuity between terminology and factual knowledge.

15.4 Glossary Identity Prevents Definition Fragmentation

Without canonical term identity, the same publisher may gradually create several definitions for what is intended to be one concept.

For example:

Glossary:
MOOP = Definition A

Plan Page:
MOOP = Definition B

FAQ:
Out-of-Pocket Maximum = Definition C

Article:
Maximum Out-of-Pocket = Definition D

Some wording variation may be legitimate.

But if the underlying meanings diverge, the publisher has created terminology drift.

A canonical term identity makes that drift easier to detect.

15.5 Definition Consistency Does Not Require Identical Wording

A publisher may explain the same concept differently for different audiences.

For example:

Canonical Definition:
formal domain definition


Consumer Explanation:
plain-language explanation


FAQ Answer:
question-specific explanation


Machine Representation:
canonical semantic definition

These representations do not need to be textually identical.

They need to remain semantically compatible.

Glossary integrity measures preservation of meaning, not repetition of wording.

15.6 Definitions Need Provenance Too

Definitions may originate from different sources.

For example:

Term:
Maximum Out-of-Pocket

Definition Source:
CMS

or:

Term:
Publisher-Specific Market Metric

Definition Source:
Publisher Methodology

or:

Term:
Legal Concept

Definition Source:
Statute / Case / Institutional Authority

The provenance model should preserve whether the publisher:

quoted or closely represented an external definition;
normalized terminology from an external source;
created a publisher-specific operational definition;
or synthesized a definition from several sources.

A definition should not borrow authority from a source that did not actually define it.

15.7 Definitions Can Be Temporal

Some definitions remain stable for decades.

Others change.

A regulatory term may be redefined.

A program definition may change by plan year.

An institutional methodology may revise the meaning of a derived metric.

Where time materially affects meaning, the glossary should preserve temporal context.

Term
        ↓
Definition Version
        ↓
Effective Period
        ↓
Source

This extends the Temporal Knowledge Integrity model from Part 14 into terminology.

15.8 Definitions Can Be Domain-Specific

The same word may mean different things in different domains.

For example:

Premium

Insurance:
amount paid for coverage

Finance:
amount above a reference value

Commerce:
higher-quality or higher-priced offering

A term identity should therefore preserve domain scope where ambiguity is material.

Likewise:

Segment

may have one meaning in Medicare plan data and a completely different meaning in marketing analytics.

A definition without domain context can create false semantic equivalence.

15.9 Glossary Relationships Matter

Terms do not exist in isolation.

A terminology system may contain relationships such as:

MOOP
→ abbreviationOf
→ Maximum Out-of-Pocket

Maximum Out-of-Pocket
→ relatedTo
→ Cost Sharing

Contract ID
→ identifies
→ Contract

Plan ID
→ identifies
→ Plan

Segment ID
→ identifies
→ Segment

Other useful relationships may include:

broaderTerm
narrowerTerm
relatedTerm
alternateLabel
supersedes
definedBy

The applicable vocabulary depends on the domain and implementation.

The important principle is that known terminology relationships should not need to be reconstructed solely from prose.

15.10 Definitions Participate in Resolution

Terminology can itself create a resolution problem.

Consider:

“What is MOOP?”

The resolution path may be:

Input:
MOOP

        ↓

Recognize:
Abbreviation

        ↓

Resolve:
term:maximum_out_of_pocket

        ↓

Domain:
Medicare Advantage

        ↓

Canonical Definition

Now consider:

“What is the MOOP for this plan?”

The term-resolution problem becomes part of a factual-resolution problem:

MOOP
        ↓
Maximum Out-of-Pocket
        ↓
Plan / Segment Identity
        ↓
Applicability
        ↓
Applicable Assertion

This demonstrates why terminology belongs inside the knowledge architecture rather than outside it as an editorial appendix.

15.11 Glossary Coverage

The first major GII dimension is Glossary Coverage.

Glossary Coverage asks:

Which material domain terms used by the publisher have canonical definitions?

For example:

Material Terms Identified:
100

Canonical Definitions:
82

Glossary Coverage:
82%

Coverage gaps may be prioritized according to:

frequency of use;
domain importance;
ambiguity risk;
regulatory significance;
participation in important assertions;
and participation in important resolution paths.

15.12 Identity Coverage

A glossary may contain definitions while still lacking stable term identity.

For example:

Glossary Entries:
100

Canonical Term IDs:
64

Identity Coverage:
64%

Identity coverage matters because definitions without stable identity are harder to reference consistently across fragments and resources.

A term that participates throughout the publisher’s knowledge system should generally be identifiable independently of the prose page on which its definition appears.

15.13 Definition Completeness

A glossary entry may exist but still be semantically incomplete.

Depending on the domain, a complete term representation may require:

canonical identity;
preferred label;
definition;
domain;
abbreviation or alternate labels where applicable;
provenance;
related terms;
temporal applicability where necessary;
and a canonical resource.

Not every field is required for every term.

The relevant question is whether enough structure exists to interpret the term correctly.

15.14 Definition Consistency

The publisher can compare uses of a term across its own resources.

For example:

Canonical:
Maximum Out-of-Pocket

Plan Resources:
compatible

County Resources:
compatible

FAQ:
compatible

Article:
compatible

One Legacy Resource:
incompatible

This produces a publisher-side consistency measurement.

The objective is not to enforce identical sentences.

It is to detect material semantic divergence.

15.15 Provenance Coverage

For terms requiring authoritative or externally sourced definitions, the publisher may measure whether provenance is recoverable.

For example:

Externally Defined Terms:
75

Terms With Recoverable Provenance:
71

Provenance Coverage:
94.7%

Publisher-created terms can likewise identify the responsible publisher or methodology where appropriate.

This makes terminology auditable rather than merely asserted.

15.16 Relationship Coverage

A mature terminology system may measure whether material relationships among terms are represented.

For example:

Known Abbreviation Relationships:
25

Represented:
25

Coverage:
100%


Known Broader/Narrower Relationships:
40

Represented:
31

Coverage:
77.5%

This is particularly useful where terms form taxonomies or where abbreviations and aliases create identity ambiguity.

15.17 Usage Coverage

A canonical glossary becomes more useful when material knowledge objects actually reference it.

For example:

Assertions Using Controlled Terms:
10,000

Assertions Referencing
Canonical Term Identity:
8,700

Usage Coverage:
87%

This measures whether the terminology system is integrated into the publisher’s knowledge architecture rather than existing as an isolated glossary.

15.18 Resolution Coverage

Query-to-Resolution Mapping may reveal terminology-specific information needs.

For example:

"What is MOOP?"

"What does SNP mean?"

"What is a Contract ID?"

"What's the difference between
a Plan ID and Contract ID?"

The publisher can test whether its terminology system supports resolution of:

preferred terms;
abbreviations;
aliases;
domain ambiguity;
term-to-entity relationships;
and related concepts.

This produces Glossary Resolution Coverage.

15.19 Publisher-Side GII Dimensions

The Glossary Integrity Index can therefore begin with publisher-controlled dimensions such as:

Dimension	Question
Coverage	Are material terms represented?
Identity	Do material terms have stable canonical identity?
Completeness	Does each term contain the semantic context required for interpretation?
Consistency	Does the term retain compatible meaning across publisher resources?
Provenance	Can material definitions be traced to their source or responsible publisher?
Relationships	Are important term relationships represented?
Usage	Do publisher knowledge objects reference canonical terminology where appropriate?
Resolution	Can important terminology information needs be resolved?

These dimensions measure the publisher’s terminology infrastructure.

They do not depend on any particular external model.

15.20 GII Should Not Hide Critical Defects

As with other WebMEM measurements, a single aggregate score can hide important weaknesses.

For example:

Glossary Integrity Index:
92 / 100

may look excellent.

But suppose:

Coverage:
100%

Identity:
100%

Completeness:
96%

Consistency:
98%

Provenance:
52%

In a regulated domain, the provenance weakness may be material despite the high aggregate score.

The GII should therefore be presented primarily as a diagnostic profile.

If an aggregate score is used, the underlying dimensions should remain visible.

15.21 Risk-Weighted Glossary Integrity

Not every term carries equal semantic risk.

For example:

Lower Risk:
general explanatory synonym

Higher Risk:
benefit category

Higher Risk:
eligibility term

Critical:
term that changes applicability
of coverage or regulation

A publisher may therefore prioritize glossary work according to:

frequency;
ambiguity;
domain importance;
resolution importance;
and consequence of misinterpretation.

This prevents a large number of trivial definitions from masking a small number of critical terminology gaps.

15.22 The Consumer Side: Definition Fidelity

The second half of glossary measurement begins after publication.

The publisher can observe how external systems interpret defined terms.

For example:

PUBLISHER REFERENCE

Term:
Maximum Out-of-Pocket

Abbreviation:
MOOP

Domain:
Medicare Advantage

Definition:
Canonical publisher definition

        ↓

CONSUMER REFLECTION

"MOOP is the most you generally pay
out of pocket for covered Medicare
Advantage services during the applicable
period, subject to the plan's terms."

The wording may differ substantially from the canonical definition.

The relevant question is:

Did the material meaning survive?

This is Definition Fidelity.

15.23 Definition Fidelity Is Not Phrase Matching

The original Glossary Impact Index measured Paraphrase Fidelity partly by how closely machine responses echoed publisher language.

The current framework makes an important correction.

A machine does not need to reproduce the publisher’s wording to preserve the publisher’s meaning.

For example:

Publisher:
"A maximum out-of-pocket limit is the
maximum amount a member may be required
to pay for applicable covered services
during the defined coverage period."

Consumer:
"The MOOP is the most you may have to
pay out of pocket for covered services
during the plan year."

may be semantically faithful even though the wording differs.

By contrast:

Consumer:
"MOOP is the most you can ever spend
on healthcare in a year."

may materially broaden the definition beyond its intended scope.

Definition Fidelity measures semantic preservation, not linguistic imitation.

15.24 Definition Fidelity Has Several Dimensions

A definition can fail in several different ways.

For example, a consumer may preserve the general concept while losing:

domain scope;
temporal scope;
exceptions;
qualifying conditions;
the distinction between related terms;
or the relationship between an abbreviation and its canonical term.

A terminology observation may therefore evaluate:

Dimension	Question
Concept Fidelity	Was the core meaning preserved?
Domain Fidelity	Was the term interpreted within the correct domain?
Scope Fidelity	Were material limits and qualifying conditions preserved?
Identity Fidelity	Was the term resolved to the correct canonical concept?
Relationship Fidelity	Were abbreviations, aliases, broader terms, or related concepts interpreted correctly?
Provenance Fidelity	Was the definition attributed appropriately where attribution was present?
Temporal Fidelity	Was the applicable definition version preserved where time matters?

This creates a much richer measurement than simple textual similarity.

15.25 Definition Errors Can Propagate Into Factual Errors

A terminology error does not necessarily remain confined to the glossary.

Consider:

Term:
MOOP

Incorrect Interpretation:
all healthcare spending

        ↓

Assertion:
MOOP = 4900 USD

        ↓

Consumer Interpretation:
Member can never spend more than
$4,900 on healthcare during the year.

The numerical value may have been reproduced perfectly.

The resulting answer is still wrong because the predicate was interpreted too broadly.

This demonstrates an important relationship:

Definition Fidelity
        ↓
Assertion Interpretation
        ↓
Factual Fidelity

If the machine misunderstands the field, reproducing the field’s value accurately does not rescue the answer.

15.26 Definition Errors Can Propagate Into Resolution Errors

Terminology can also affect resolution.

Consider:

Input:
Contract ID

Consumer Interpretation:
Plan ID

The terminology error may cause the consumer to resolve:

H5521
→ Plan

instead of:

H5521
→ Contract

Every subsequent relationship may then be traversed incorrectly.

Likewise:

Segment
→ interpreted as Plan

can cause segment-specific benefits to be generalized to the parent Plan.

This makes glossary integrity part of resolution integrity.

15.27 Definition Queries Provide Direct Test Cases

Some glossary fidelity tests can begin with direct definition queries.

For example:

"What is MOOP?"

"What does SNP mean in Medicare?"

"What is a CMS Contract ID?"

"What is a Medicare Plan ID?"

"What is a Segment ID?"

The observed output can then be compared against the canonical DefinedTermFragment.

These tests may evaluate:

term identity;
preferred meaning;
domain;
abbreviation expansion;
scope;
related concepts;
and provenance where material.

15.28 Definitions Should Also Be Tested in Context

Direct definition queries are useful but incomplete.

A consumer may define a term correctly when asked directly and still misuse it inside another information need.

For example:

Direct Query:
"What is a Medicare Segment?"

Result:
correct


Applied Query:
"What is the premium for Plan X?"

Result:
Segment-specific premium generalized
to the entire Plan

The definition was understood in isolation.

Its semantic role was not preserved during resolution.

A mature glossary measurement program should therefore test both:

DEFINITION FIDELITY

Can the consumer explain the term?


APPLIED TERM FIDELITY

Does the consumer use the term's
meaning correctly during resolution?

15.29 Applied Term Fidelity

Applied Term Fidelity measures whether a defined concept retains its meaning when used inside a larger answer or resolution path.

For example:

Canonical Term:
Segment

Meaning:
Geographically applicable subdivision
of a Plan within the represented domain

An applied test might ask:

“Does this Plan have the same premium in every county?”

The consumer’s answer can reveal whether it understands that Segment identity may determine geographic variation.

Applied Term Fidelity therefore connects glossary measurement to:

identity fidelity;
relationship fidelity;
applicability fidelity;
and resolution fidelity.

15.30 Attribution Is Separate From Definition Fidelity

A consumer may reproduce a definition accurately without citing the publisher.

For example:

Definition Fidelity:
PASS

Citation Present:
NO

Another consumer may cite the publisher while materially distorting the definition:

Definition Fidelity:
FAIL

Citation Present:
YES

These are different outcomes.

The revised GII should therefore avoid combining citation and definition accuracy into one concept such as “citation confidence.”

Measure each independently.

15.31 Consumer-Specific Definition Profiles

Part 13 established that machine consumers should be observed independently against one publisher reference.

The same model applies to glossary terms.

For example:

Canonical Term:
Maximum Out-of-Pocket

        ↓
   ┌────┼────┬────┐
   ↓    ↓    ↓    ↓
   A    B    C    D

Consumer A:
Concept Fidelity = PASS
Scope Fidelity = PASS

Consumer B:
Concept Fidelity = PASS
Scope Fidelity = FAIL

Consumer C:
Identity Fidelity = FAIL

Consumer D:
Concept Fidelity = PASS
Citation Present = YES

The publisher maintains one canonical definition.

Each consumer produces a different reflection.

One term identity. One publisher definition. Many observed interpretations.

15.32 Cross-Consumer Definition Fidelity

A publisher may measure definition fidelity across a defined set of consumers.

For example:

Term	Consumer A	Consumer B	Consumer C	Consumer D
MOOP	98%	94%	81%	96%
Contract ID	100%	91%	87%	98%
Segment	93%	72%	68%	90%

Such measurements require a defined observation method and actual evidence.

They describe observed fidelity during the applicable period.

They are not permanent properties of the consumers.

15.33 Terminology Drift Over Time

Definition fidelity can also change over time.

For example:

June:
Segment definition correct

July:
Segment definition correct

August:
Segment conflated with Plan

September:
Segment definition correct again

The publisher can describe this as observed terminology drift.

It should not automatically conclude:

The consumer forgot the definition
and later relearned it.

The internal mechanism remains unknown.

This applies the Temporal Knowledge Integrity model to glossary observation.

15.34 Publisher Terminology Drift

The publisher’s own terminology can drift as well.

For example:

Canonical Glossary:
Segment = Definition A

Plan Template:
Segment = Definition A

Legacy Article:
Segment = Definition B

or:

Glossary:
Contract ID identifies Contract

FAQ:
Contract ID described as Plan ID

These are publisher-side integrity failures.

They should be corrected independently of how external machines currently behave.

15.35 Glossary Integrity and Cross-Surface Consistency

Part 8 established that publisher-controlled surfaces should remain semantically compatible.

Glossary terms provide one of the clearest cross-surface test cases.

A publisher may compare:

Canonical DefinedTermFragment
        ↕
Glossary Page
        ↕
FAQ
        ↕
Plan Resources
        ↕
Resolver Resources
        ↕
Articles
        ↕
Human Interface Labels

The wording may vary.

The canonical meaning should remain stable.

15.36 Glossary Integrity and Query-to-Resolution Mapping

Part 10 established that recurring information needs reveal resolution requirements.

Terminology questions reveal where definition architecture matters.

For example:

Queries:

"What is H5521?"

"What does Contract ID mean?"

"Is H5521 a Plan ID?"

"What's the difference between
a Contract ID and Plan ID?"

        ↓

Terminology Requirement:

Contract ID
→ identifies
→ Contract

Plan ID
→ identifies
→ Plan

Contract
→ hasPlan
→ Plan

The glossary does more than define the words.

It helps establish the semantic distinctions required for resolution.

15.37 Glossary Integrity and Representation Optimization

Part 11 established Representation Optimization as improvement of publisher-controlled semantic structure.

Glossary optimization may therefore include:

adding canonical term identities;
resolving duplicate concepts;
adding missing provenance;
clarifying domain scope;
representing abbreviations and aliases;
adding important term relationships;
correcting cross-surface terminology drift;
integrating terms into applicable assertions;
and improving resolution coverage for terminology queries.

The objective is not to make the machine repeat the glossary more often.

The objective is to make the publisher’s vocabulary more coherent.

15.38 The GII Has Two Sides

The revised Glossary Integrity Index should preserve the boundary established throughout this paper.

It therefore has two distinct measurement surfaces:

PUBLISHER SIDE

Glossary Coverage
Identity Coverage
Definition Completeness
Definition Consistency
Provenance Coverage
Relationship Coverage
Usage Coverage
Resolution Coverage


────────────────────────────
      MACHINE BOUNDARY
────────────────────────────


OBSERVABLE CONSUMER SIDE

Definition Presence
Concept Fidelity
Domain Fidelity
Scope Fidelity
Identity Fidelity
Relationship Fidelity
Provenance Fidelity
Temporal Fidelity
Applied Term Fidelity

The publisher controls the first.

The publisher observes the second.

They should not be collapsed into one causal claim.

15.39 A Glossary Integrity Dashboard

A practical publisher-side dashboard might show:

GLOSSARY INTEGRITY

Material Terms Identified          120
Canonical Terms                    114

Coverage                           95%
Identity Coverage                  98%
Definition Completeness            94%
Cross-Surface Consistency          97%
Provenance Coverage                91%
Relationship Coverage              82%
Usage Coverage                     88%
Resolution Coverage                93%

A separate consumer-observation view might show:

OBSERVED DEFINITION FIDELITY

Definition Presence                78%
Concept Fidelity                   96%
Domain Fidelity                    94%
Scope Fidelity                     87%
Identity Fidelity                  92%
Relationship Fidelity              85%
Provenance Fidelity                81%
Applied Term Fidelity              84%

These views answer different questions.

The first asks:

How strong is our terminology system?

The second asks:

How faithfully is that terminology
being reflected externally?

15.40 Do Not Collapse the GII Too Early

It may be tempting to calculate:

GII = 91.4

and use that number as the primary result.

That should be approached cautiously.

Suppose:

Coverage:
100%

Consistency:
100%

Identity:
100%

Provenance:
45%

A high average could conceal the provenance weakness.

Or:

Definition Fidelity:
98%

Applied Term Fidelity:
52%

The machine can define the terminology correctly but fails to use it correctly during resolution.

That distinction is far more useful than the average.

The GII should function first as a diagnostic profile and only secondarily, if at all, as an aggregate score.

15.41 Risk-Weighted GII

If an aggregate measure is operationally useful, weighting should reflect domain risk rather than treating every term equally.

For example:

Term:
Marketing Description

Risk:
LOW


Term:
MOOP

Risk:
HIGH


Term:
Eligibility Category

Risk:
CRITICAL

A missing synonym for a low-risk explanatory term should not offset a serious definition defect in a term governing eligibility or applicability.

Risk weighting can help prioritize remediation while preserving the underlying dimensional measurements.

15.42 Glossary Priority Mapping

The publisher can also prioritize terms without calculating one universal GII score.

For example:

Term	Domain Risk	Publisher Integrity	Observed Fidelity	Priority
MOOP	High	High	Medium	Observe / investigate
Segment	High	Medium	Low	Immediate review
Contract ID	High	High	High	Maintain
General Synonym	Low	Medium	Medium	Low priority

This kind of priority map is more actionable than a leaderboard of glossary terms.

It identifies where publisher weakness, observed machine weakness, and domain consequence intersect.

15.43 Publisher Integrity and Consumer Fidelity Must Be Diagnosed Together

The two sides of the GII become especially useful when examined together.

Consider four possible states:

Publisher Integrity	Observed Fidelity	Interpretation
High	High	Definition is well represented and generally survives observed machine interpretation
High	Low	Publisher representation appears sound; investigate consumer behavior before changing it
Low	High	Machines may currently compensate for publisher weakness; representation should still be improved
Low	Low	Publisher-side terminology weakness is a strong remediation candidate

This is the same reference-and-reflection model established throughout the revised framework.

Canonical DefinedTerm
        ↓
Publisher Reference
        ↓
────────────────────────
     MACHINE BOUNDARY
────────────────────────
        ↓
Consumer Interpretation
        ↓
Observed Reflection

The glossary gives the publisher an unusually clean semantic reference because the expected meaning is explicitly defined.

15.44 High Publisher Integrity Does Not Guarantee High Definition Fidelity

A publisher may construct an excellent terminology system and still observe poor external interpretation.

For example:

Canonical Identity:
PASS

Definition:
PASS

Domain:
PASS

Provenance:
PASS

Relationships:
PASS

Cross-Surface Consistency:
PASS

Observed Definition Fidelity:
FAIL

The publisher should investigate the result.

But it should not automatically rewrite a correct definition merely to accommodate one consumer’s incorrect reflection.

Do not optimize correct terminology toward a machine’s misunderstanding of it.

15.45 High Definition Fidelity Does Not Excuse Weak Publisher Integrity

The inverse is equally important.

A machine may correctly define a term even when the publisher’s own terminology architecture is poor.

For example:

Publisher:
ambiguous definition

Consumer:
correct definition from another source

The observed answer may be excellent.

That does not mean the publisher representation is excellent.

The consumer may have resolved the meaning through:

another publisher;
an authoritative external source;
training knowledge;
a knowledge graph;
or another retrieval path.

A correct reflection does not repair a deficient reference.

15.46 Glossary Observation Should Use Query Families

Definition testing should not depend on one exact prompt.

For example, the concept:

term:maximum_out_of_pocket

might be tested through:

"What is MOOP?"

"What does MOOP mean in Medicare?"

"What is a maximum out-of-pocket limit?"

"What does out-of-pocket maximum mean
for a Medicare Advantage plan?"

These queries belong to a terminology query family.

The publisher can evaluate whether the same canonical concept survives different natural-language expressions.

This tests resolution robustness without requiring exact phrase matching.

15.47 Ambiguous Terms Require Contextual Tests

Some terms cannot be evaluated properly without domain context.

For example:

"What is a premium?"

is broader than:

"What is a monthly plan premium
in Medicare Advantage?"

Likewise:

"What is a segment?"

may be ambiguous across many domains.

A glossary test should distinguish:

General-Language Interpretation

from

Domain-Scoped Interpretation

A consumer should not necessarily fail because it provides a valid general definition when the query supplied no domain context.

Measurement should test the information need actually presented.

15.48 Glossary Resolution Can Require Clarification

A term may have several legitimate meanings.

For example:

Term:
Premium

Possible Domains:
Insurance
Finance
Commerce

If the query supplies no domain context, the correct resolution may remain ambiguous.

A consuming system may:

provide several meanings;
ask for clarification;
or infer a domain from established conversational context.

The publisher should not pretend that one domain-specific definition is universally applicable.

Glossary integrity includes preserving legitimate lexical ambiguity.

15.49 Definitions Can Form Resolution Networks

A sophisticated terminology system may become more than a flat list.

For example:

Medicare Advantage
        ↓
hasConcept
        ↓
Maximum Out-of-Pocket
        ↓
abbreviation
        ↓
MOOP

Maximum Out-of-Pocket
        ↓
relatedTo
        ↓
Cost Sharing

Cost Sharing
        ↓
includesConcept
        ↓
Copayment

Cost Sharing
        ↓
includesConcept
        ↓
Coinsurance

This structure allows machines to move among related concepts without requiring every glossary entry to reproduce the definitions of all neighboring terms.

The glossary becomes a terminology resolution network.

15.50 Defined Terms Can Connect to Non-Glossary Knowledge

Term identity becomes especially useful when definitions connect to factual or procedural knowledge elsewhere in the publisher’s system.

For example:

Defined Term:
Maximum Out-of-Pocket

        ↓

Used By:
Plan Benefit Assertions

        ↓

Referenced In:
County Comparisons

        ↓

Explained By:
Human Glossary Resource

or:

Defined Term:
Special Enrollment Period

        ↓

Related Policy:
Enrollment Rules

        ↓

EligibilityFragment:
Qualifying Conditions

        ↓

ProcedureFragment:
Enrollment Process

The glossary therefore helps bind vocabulary to the knowledge structures that use it.

15.51 Glossary Integrity Supports Semantic Recoverability

Semantic recoverability depends partly on whether a machine can recover the meaning of the predicates and concepts used throughout the publisher’s knowledge.

Consider:

maximum_out_of_pocket = 4900 USD

If the field meaning is unknown, the value is only partially recoverable.

With terminology support:

Assertion:
maximum_out_of_pocket = 4900 USD

        ↓

Defined Term:
Maximum Out-of-Pocket

        ↓

Domain:
Medicare Advantage

        ↓

Canonical Definition:
...

the assertion carries a clearer semantic path.

Glossary architecture helps machines recover not merely the value, but what the value means.

15.52 Glossary Integrity Supports Consumer Independence

A canonical terminology system also supports the consumer-independence principle established in Part 13.

The publisher should not maintain:

MOOP definition for Consumer A

MOOP definition for Consumer B

MOOP definition for Consumer C

merely because the consumers currently paraphrase the concept differently.

The publisher maintains:

Canonical Publisher Definition
        ↓
Consumer A Reflection
Consumer B Reflection
Consumer C Reflection

The reflections are measured against one reference.

Definitions belong to the publisher’s knowledge model, not to the consumer’s preferences.

15.53 Glossary Integrity Supports Temporal Knowledge Integrity

Where definitions change over time, term identity should preserve the relationship between versions.

For example:

Definition Version A
effective through 2026-12-31

        ↓
supersededBy
        ↓

Definition Version B
effective beginning 2027-01-01

Historical resources may continue to require Version A.

Current resources may require Version B.

The term identity may remain stable while the applicable definition version changes.

This prevents current terminology from silently rewriting historical meaning.

15.54 Glossary Integrity Supports Provenance

Defined terms also strengthen provenance by clarifying what source language means.

Suppose a source dataset contains:

MOOP_AMT = 4900

The publisher may normalize that field into:

maximum_out_of_pocket = 4900 USD

The provenance chain can preserve:

Source Field:
MOOP_AMT

        ↓
Publisher Normalization

Canonical Field:
maximum_out_of_pocket

        ↓
Defined Term:
Maximum Out-of-Pocket

This makes the transformation from source vocabulary to publisher vocabulary inspectable.

15.55 Glossary Integrity Supports Data Normalization

This is especially valuable when multiple source datasets use different labels for equivalent or related concepts.

For example:

Source A:
MOOP_AMT

Source B:
MAX_OOP

Source C:
OUT_OF_POCKET_LIMIT

        ↓

Publisher Normalization

        ↓

Canonical Concept:
Maximum Out-of-Pocket

The DefinedTerm identity can provide a stable semantic target for the normalized publisher knowledge.

This does not imply that differently named source fields are automatically equivalent.

The publisher must establish equivalence based on domain knowledge.

15.56 The Glossary Can Reveal Data-Model Problems

Terminology work often exposes deeper inconsistencies in the publisher’s data architecture.

For example:

Database A:
Plan

Database B:
Product

Template:
Plan

Analytics:
Policy

If all four labels refer to the same entity type, normalization may be needed.

If they refer to materially different concepts, collapsing them would be incorrect.

The glossary forces the publisher to answer:

Do these words mean the same thing?

That question can expose hidden modeling problems long before they appear in machine-generated answers.

15.57 The Glossary Is a Governance Surface

Because terminology participates throughout the knowledge system, changes to important definitions may require governance.

A mature publisher may track:

who owns a term;
who may approve changes;
the source supporting the definition;
the effective date;
affected resources;
affected assertions;
and downstream resolution paths.

For example:

Definition Change
        ↓
Identify Dependent Assertions
        ↓
Identify Dependent Resources
        ↓
Validate Semantic Impact
        ↓
Publish Updated Definition
        ↓
Validate Cross-Surface Consistency

This is terminology governance, not glossary maintenance in the traditional editorial sense.

15.58 Glossary Changes Can Have Large Blast Radius

A small wording change may be semantically insignificant.

A genuine definition change may affect large portions of the publisher’s knowledge system.

For example:

Defined Term
        ↓
Used By 4,000 Assertions
        ↓
Referenced By 2,000 Resources
        ↓
Participates In 12 Resolver Profiles

A material definition change therefore deserves dependency analysis.

The terminology layer can become one of the most connected parts of the publisher’s knowledge architecture.

15.59 Glossary Integrity Is Not Glossary Size

A large glossary is not necessarily a strong glossary.

For example:

Publisher A:
5,000 loosely defined terms

Publisher B:
300 canonical, provenanced,
well-related domain terms

Publisher B may have the stronger terminology architecture.

The objective is not maximum term count.

It is sufficient coverage of the concepts material to the publisher’s knowledge and resolution responsibilities.

15.60 Not Every Word Needs a DefinedTermFragment

The protocol should not encourage publishers to turn ordinary language into unnecessary semantic objects.

A term is a stronger candidate for explicit representation when it:

has domain-specific meaning;
is materially ambiguous;
appears repeatedly in structured assertions;
participates in resolution;
has regulatory or operational significance;
has important abbreviations or aliases;
or requires provenance or versioning.

Ordinary words that carry no special semantic responsibility may remain ordinary words.

Terminology engineering should reduce ambiguity, not manufacture complexity.

15.61 GII Can Prioritize Terminology Work

The revised GII becomes particularly useful as a prioritization system.

Suppose:

Term A

Domain Risk:
HIGH

Publisher Integrity:
LOW

Observed Fidelity:
LOW

Priority:
CRITICAL


Term B

Domain Risk:
HIGH

Publisher Integrity:
HIGH

Observed Fidelity:
LOW

Priority:
INVESTIGATE


Term C

Domain Risk:
LOW

Publisher Integrity:
LOW

Observed Fidelity:
HIGH

Priority:
LOW


Term D

Domain Risk:
HIGH

Publisher Integrity:
HIGH

Observed Fidelity:
HIGH

Priority:
MAINTAIN

This gives the publisher a rational way to allocate terminology-engineering effort.

15.62 The GII Becomes a Diagnostic Instrument

The original Glossary Impact Index attempted to answer:

How strongly does the machine
remember this definition?

The revised Glossary Integrity Index asks several more useful questions:

Have we defined the term?

Have we identified it?

Have we scoped it?

Have we sourced it?

Have we related it?

Do our own surfaces agree about it?

Do our assertions use it consistently?

Can important terminology queries
resolve through it?

When machines reflect the term,
does its meaning survive?

When they use the term in context,
do they apply its meaning correctly?

Every publisher-side question can be investigated directly.

Every consumer-side question can be observed without claiming access to hidden model state.

15.63 From Memory Weight to Semantic Stability

The evolution of the GII can now be stated clearly:

GLOSSARY IMPACT INDEX — 2025

Definition
        ↓
Multi-Format Exposure
        ↓
Reinforcement
        ↓
Machine Recall
        ↓
Memory Weight
        ↓
Impact Score


GLOSSARY INTEGRITY INDEX — 2026

Term Identity
        ↓
Definition
        ↓
Domain
        ↓
Provenance
        ↓
Relationships
        ↓
Publisher Usage
        ↓
Resolution Coverage
        ↓
Cross-Surface Consistency
        ↓
Published Reference
        ↓
Observed Definition Fidelity

The first attempted to score something inside the machine.

The second evaluates the stability of meaning across the publisher-consumer boundary.

15.64 The Glossary Is Not an Appendix

One claim from the original framework survives almost unchanged:

The glossary is not an appendix.

But the reason is clearer now.

The glossary is not important because repeated definitions become memory anchors.

It is important because terminology sits underneath the knowledge model.

Definitions determine what predicates mean.

Definitions distinguish related entity types.

Definitions establish domain scope.

Definitions help normalize source vocabulary.

Definitions participate in resolution.

Definitions can carry provenance.

Definitions can change over time.

And definitions can determine whether a perfectly reproduced factual value is interpreted correctly.

That makes the terminology layer infrastructure.

15.65 The New Glossary Principle

The revised framework can reduce glossary architecture to a simple principle:

If a term materially changes how published knowledge should be interpreted, its meaning should not be left unnecessarily implicit.

The publisher should establish:

What does this term mean?

What concept does it identify?

Within what domain?

According to whom?

How does it relate to neighboring concepts?

Where is it used?

When does this definition apply?

Then external systems can be measured against that explicit reference.

Conclusion

The original Glossary Impact Index attempted to measure which definitions machines retrieved, reinforced, and remembered most strongly.

The revised Glossary Integrity Index measures something different.

It asks whether the publisher has built a terminology system capable of preserving meaning—and whether that meaning survives when external machines interpret and apply it.

On the publisher side:

Term
        ↓
Identity
        ↓
Definition
        ↓
Domain
        ↓
Provenance
        ↓
Relationships
        ↓
Usage
        ↓
Resolution
        ↓
Cross-Surface Consistency

On the observable consumer side:

Published Definition
        ↓
Machine Interpretation
        ↓
Concept Fidelity
        ↓
Domain Fidelity
        ↓
Scope Fidelity
        ↓
Identity Fidelity
        ↓
Applied Term Fidelity

The two sides meet at the publisher’s canonical definition.

That definition becomes the reference.

The machine’s interpretation becomes the reflection.

The difference between them becomes measurable.

This gives the glossary a more important role than the original framework assigned to it.

It is not merely a collection of definitions intended to improve recall.

It is part of the semantic infrastructure through which the rest of the publisher’s knowledge becomes interpretable.

A factual assertion depends on understanding its predicate.

An entity relationship depends on understanding the relationship being expressed.

A Resolver depends on distinguishing the entities and concepts through which resolution proceeds.

A policy depends on understanding the terms that define its conditions.

An eligibility rule depends on understanding the categories to which it applies.

When terminology fails, failures can propagate throughout the knowledge system.

Definition Error
        ↓
Interpretation Error
        ↓
Relationship Error
        ↓
Applicability Error
        ↓
Resolution Error
        ↓
Wrong Answer

This is why glossary integrity deserves independent measurement.

The objective is not:

Make machines remember
our definitions.

It is:

Make our definitions explicit,
stable, attributable, connected,
and semantically recoverable.

Then observe whether machines preserve that meaning.

The evolution from the original framework can therefore be reduced to:

2025

Definition
→ Reinforcement
→ Recall
→ Memory Weight


2026

Definition
→ Identity
→ Meaning
→ Provenance
→ Relationships
→ Usage
→ Resolution
→ Published Reference
→ Observed Fidelity

Your definitions do not need to control machine memory.

They need to control what your own vocabulary means.

That is the purpose of the Glossary Integrity Index.