Appendix A

Glossary, Source Guide, and Research Traceability

Why this guide exists

AIOS introduces familiar words—memory, context, agent, workflow, sovereignty, knowledge, intelligence—but uses them within a specific architecture. This guide gives each major term one stable meaning and makes the route from public claims to underlying research legible.

It is not a hidden rulebook. The canonical explanations remain in the chapters where the mechanisms are developed. The glossary provides compact orientation; cross-links provide exact descent.

The most important reading principle is:

Similar words do not imply identical mechanisms, and a durable record does not acquire authority merely because it exists.

Concept ownership map

The glossary summarizes terms; it does not govern them independently of the chapters where their mechanisms are developed.

Concept family	Owning chapters
Whole-system thesis and emergent intelligence	Core Thesis, Research Overview, From Model Intelligence to System Intelligence, and Emergent Intelligence and the AI-Native Paradigm
Intent engineering, purpose traceability, and productized intent	Emergent Intelligence and the AI-Native Paradigm, Bottom-Up Personal Composition, Top-Down Operational Knowledge, and Planning
Fractal Seed, semantic situation, and cognitive movements	Fractal Seed, Semantic Situation, and Cognitive Movements
Context, memory, files, and document composition	Context Composition, Continuous Context Optimization, Memory Ecology, Files and Metadata, and Document Hierarchy
Bottom-up formation and top-down operational knowledge	Bottom-Up Personal Composition and Top-Down Operational Knowledge
Workflows, menus, plans, agency, delegation, and authority	Workflows, Semantic Action Fields, Planning, Asynchronous Agency, Distributed Cognition, and Bounded Semantic Sovereignty
Promotion, Atlas, correction, mechanics, and integrated journeys	Registration and Promotion, Flow Atlas, Self-Correction, Product Mechanics, and Reference Journeys
Evaluation, economics, and research positioning	Capability Horizons, Economic Architecture, and Research Positioning
Visual translation and open research	Designer Visual System and Open Research Questions

Foundational concepts

Term	Meaning in AIOS
AIOS	The local-first, person- or organization-governed intelligence system described in this overview. It composes context, models, files, cognitive movements, exact operations, and reintegration into a continuing reasoning environment.
General intelligence harness	The complete scaffolding through which general model capability becomes situated, bounded, persistent, and useful across many kinds of cognitive work. It is broader than an agent loop or prompt wrapper.
AI-native architecture	Software architecture that treats probabilistic semantic judgment as a first-class capability inside explicit contracts for context, authority, artifacts, exact effects, verification, and reintegration. Adding chat or function calling to a conventional application is not sufficient by itself.
System intelligence	Useful capability attributable to the complete person–model–artifact–operation environment rather than to the model alone.
System-level capability	The empirically observed performance of a complete model–harness–environment configuration across a declared task distribution, resource budget, and time horizon. It must be evaluated end to end rather than inferred from model scores alone.
Emergent intelligence	The AIOS architectural thesis that system-level capability depends on organized interactions among components and cannot be assigned to one operational center across the intended task distribution. It does not imply consciousness or strong metaphysical emergence.
Inference-burden transfer	The hypothesis that durable purpose, relationships, decisions, provenance, and work state move recurring reconstructive work out of live model inference, reserving inference for what is genuinely new.
Cumulative capability amplification	The hypothesis that multiple bounded improvements to context, memory, cognitive fit, authority, verification, and reintegration can interact to produce a larger whole-system gain. It does not assume that all gains are additive or monotonic.
Hallucination-surface reduction	The hypothesis that explicit sources, standing, meaningful absence, narrower jurisdictions, bounded outputs, and verification reduce the occasions on which a model must invent unsupported ground. It does not imply elimination of model error.
Accepted outcome	A result that adequately serves its purpose, meets relevant semantic and evidential standards, respects authority, produces only permitted effects, and is coherently integrated into subsequent work.
Personal AGI	A locally governed general reasoning system built around a person’s durable knowledge, judgments, standards, and purposes. It describes an architectural and research direction, not a claim that one local model satisfies a universal AGI test.
Self-contained domain knowledge system	A locally intelligible domain—such as professional work, research, education, cooking, health, or an organization—with enough purpose, terminology, evidence, artifacts, decisions, methods, relationships, and history to remain usable without hidden provider memory. Self-contained does not mean isolated from outside models or information.
Internally consistent	Organized so that terms, standing, relationships, disagreements, decisions, and changes remain legible across the domain. Internal consistency permits contradiction and plural views; it does not mean unanimity or truth by coherence alone.
Intent engineering	The development, preservation, mapping, and revision of purpose as governing ground for later cognitive movements. It can begin with a tension or half-formed aim; human authority determines what purpose becomes committed.
Productized intent	Purpose and expertise made durable and operational through a locally governed knowledge system: its roles, workflows, document structures, accepted practices, relationships, completion criteria, and revision paths. Productizing intent standardizes recurring work without making every case identical or transferring human authority to the system.
Purpose traceability	The ability to explain how present work serves a governing purpose across outcomes, strategy, architecture, commitments, foreground movements, artifacts, and effects—and how results can responsibly revise that purpose.

The reasoning grammar

Term	Meaning in AIOS
Fractal Seed	The recurring Why → How → What grammar through which purpose develops into a result that can be reintegrated and renew subsequent ground. “Fractal” denotes functional recurrence across scales, not a mathematically established fractal.
Why	Purpose, bearing, intended change, beneficiary, governing question, or criterion of relevance.
How	Reasoning, relationship, method, composition, plan, or transformation through which the Why develops.
What	The judgment, expression, artifact, decision, effect, unresolved condition, or possibility produced by the movement.
Result	The semantic product carried by the What and evaluated against the purpose that generated it.
Reintegration	The judgment through which a result changes—or explicitly does not change—the surrounding artifact, plan, decision field, memory, or knowledge system.
Renewal	The next semantic situation formed after reintegration. It is the new ground from which another Why can be understood.
Situation renewal	The standing-aware transition by which a movement's relevant semantic product and accepted delta change—or explicitly leave unchanged—the active semantic situation from which subsequent context is composed.
Semantic situation	The present relationship among direction, epistemic ground, intentional ground, productive ground, and temporal or attentional ground. It is the unit AIOS reconstructs for continuing work.
Semantic altitude	The scale at which a judgment operates: passage, section, document, project, domain, or another meaningful level.
Semantic product	What a cognitive movement makes: a finding, interpretation, draft, plan, decision proposal, review, or other meaningful result.
Semantic delta	What the product confirms, qualifies, challenges, supersedes, proposes, opens, closes, or leaves unresolved in the receiving situation.

Cognitive movements and carriers

Term	Meaning in AIOS
Cognitive movement	A bounded kind of intellectual transformation, such as orienting, discovering, deciding, designing work, composing, revising, testing, integrating, or reflecting. It is the primary unit of context composition, model routing, authority, and evaluation.
Judgment form	The general type of transformation a cognitive movement performs.
Seat	One executable model-facing responsibility with a coherent context, judgment, output contract, consumer, and failure behavior. A seat is not necessarily a permanent agent.
Reasoning-turn envelope	The three logical functions around a consequential movement: attention formation, bounded situated judgment, and landing with integration and implication. The functions do not require three model calls and do not imply that every result is accepted or executed.
Attention formation	The cognitive and compositional work that establishes the present purpose, object, semantic situation, judgment form, sufficient altitude, relevant ground, and authority boundary before or around a bounded judgment. It may combine person direction, deterministic assembly, and semantic orientation.
Named consumer	The person, later judgment, artifact, workflow stage, plan, effect owner, or integration boundary able to use a movement's product. A staged judgment without a real consumer is orchestration overhead.
Role	Domain or professional expertise and standpoint composed into a movement, such as editor, strategist, researcher, or evidence reviewer.
Carrier	A replaceable form through which a movement occurs or travels: a turn, seat, workflow stage, role, delegated branch, command, file, or return.
Workflow	A reusable composition of dependent cognitive movements and quality logic for a class of outcomes. It guides judgment rather than precomputing conclusions.
Stage	One bounded contribution within a workflow, with relevant ground, a method, a product, a quality condition, and a bridge to what follows.
Agent or delegated branch	A bounded line of work permitted to proceed independently under a commission. It does not become the center of the knowledge system.
Command	A person-invoked, named entry point into a bounded capability.
Operation	An exact mechanical effect performed under validated identity, authority, and current-state conditions.

Context and evidence

Term	Meaning in AIOS
Context composition	The purposeful selection, ordering, attribution, and resolution of the semantic ground required for one cognitive movement.
Context engineering	The wider architecture by which relevant ground is authored, preserved, selected, transformed, and presented to a model. In AIOS, its central practice is content composition.
Continuous context optimization	Continual, use-shaped recomposition of live context from durable ground as purpose, state, and attention change. It is a governing design objective, not a claim that every context is globally optimal.
Situation brief	A compact, use-shaped projection of the semantic situation for a particular judgment. It is reconstructable from canonical sources and standing-aware durable records and is not itself the complete truth.
Context capsule	The exact provider-visible package for one invocation: purpose, object, sources, standing, guidance, authority boundary, requested transformation, and expected product.
Operation-aware context	Context composed according to the transformation required—such as lookup, comparison, aggregation, contradiction resolution, temporal update, structural reasoning, creation, or verification—not only the topic.
Evidence budget	The amount and resolution of evidence a movement can use effectively under its model, operation, latency, and risk conditions. It is not merely a token limit.
Provider-visible ground	The exact material sent to a model for an invocation. It is a subset or projection of the durable system.
Context warrant	An inspectable account of why a context region was included, where it came from, what standing it has, what was compressed or omitted, and when deeper ground should re-enter.
Coarsest sufficient semantic altitude	The highest-level representation that can support the intended judgment and consequence without hiding a material distinction. A named evidence, structural, wording, or mutation need justifies descent; lower-level findings are interpreted again at the receiving altitude.
Source descent	The route from a summary, relationship, or public claim back to exact supporting material.
Meaningful absence	An explicit recognition that relevant ground does not exist, is unavailable, or remains unknown.
Deliberate compression	Purpose-specific reduction that preserves standing, disagreement, rationale, uncertainty, and a route back to exact sources.
Emergency compression	Reactive shortening of an overgrown transcript without a principled account of relevance or loss. AIOS is designed to make this unnecessary as the normal continuity mechanism.

Memory and epistemic standing

Term	Meaning in AIOS
Memory ecology	Multiple forms of memory with different timescales, resolutions, authority, retrieval behavior, and forgetting conditions.
Short-term context	What is immediately present in one movement or exchange.
Working memory	The active semantic situation maintained across a related sequence of movements.
Durable memory	Files, sources, decisions, relationships, plans, artifacts, lineage, and accepted syntheses that can reconstruct relevant ground later.
Relational memory	Memory organized through inspectable relationships among ideas, artifacts, evidence, decisions, people, workflows, and time.
Recall ladder	Progressive movement from recognition or metadata, through summaries and relationships, to exact passages and complete primary sources.
Semantic standing	The umbrella for what a meaning-bearing item is permitted to claim or direct in later reasoning. It includes epistemic standing and intentional standing without collapsing evidence, belief, decision, and authorization.
Epistemic standing	An item’s relation to knowledge and evidence: source evidence, interpretation, assumption, proposed ground, accepted ground, contested ground, unknown, or superseded ground.
Intentional standing	An item’s relation to action and commitment: proposed decision, authorized commitment, active commitment, deferred commitment, revoked commitment, or superseded commitment.
Provenance	The origin and transformation history of a claim, artifact, relationship, or effect. Provenance establishes lineage and auditability; it does not establish truth.
Observed content	Material encountered or produced but not yet proposed as durable knowledge.
Proposed memory or relationship	A candidate addition awaiting review, scope, and appropriate authority.
Accepted derived knowledge	A reviewed synthesis, relationship, or method that may guide future work within its declared scope while retaining source descent.
Canonical source or artifact	The authoritative person- or organization-controlled object and revision for a declared identity: the source, document, plan, decision record, workflow, or other durable work to which later references must resolve. Canonical location does not by itself establish truth or epistemic acceptance.
Accepted ground	A claim, decision, method, or other meaning-bearing object granted epistemic or intentional standing within a declared scope by the appropriate authority. When acceptance changes durable ground, that standing settles only after the authorized write is verified and reconstructable. Acceptance does not make the object infallible or erase its source descent and reopen conditions.
Temporal state transition	A declared change such as append, correct, qualify, supersede, revoke, or preserve for history.
Promotion	The governed movement by which candidate knowledge or practice gains accepted reusable standing.
Revalidation	Later evidence that accepted knowledge or practice remains useful and valid within its scope.
Forgetting	Deliberate release from routine activation, which may mean dormancy, archival preservation, or authorized deletion rather than erasure by default.
Semantic capital	Net-positive durable structure—sources, decisions, methods, artifacts, relationships, and correction history—that lowers future reconstruction or error cost.
Semantic debt	Maintenance, confusion, error, contradiction, staleness, or governance burden created by poorly curated durable structure.

Files, metadata, and derived views

Term	Meaning in AIOS
Canonical artifact	The person- or organization-controlled file containing the primary human-meaningful object: a document, source, plan, workflow, decision, or other durable work.
Companion metadata	A file-adjacent record of identity, purpose, standing, structure, relationships, state, lineage, and reconstruction cues that would degrade the primary artifact if embedded directly.
Annotation	An anchored, provenance-bearing judgment about how an artifact, source, passage, or movement should be understood later. It may preserve interpretation, qualification, critique, relationship, contradiction, or proposed supersession without rewriting the underlying artifact or accepted spine; by itself it does not change standing, grant authority, or trigger an effect.
Document Home	A canonical companion metadata record that allows a person or model to understand a document’s purpose, structure, state, relationships, and open questions before reading all of it. Derived maps may project this record but do not outrank it.
Document Spine	The accepted compact Why–How–What account of why a document exists, how its load-bearing movements develop, and what it establishes or leaves open. It lives in companion metadata and can be projected as a Document Map without imposing a visible prose template.
Project Home	A canonical navigational record for a project, carrying a compact account of purpose, plan, artifact field, decisions, branches, returns, and foreground. It does not replace underlying artifacts, sources, or decisions.
File-native knowledge graph	The network that emerges from relationships declared beside canonical files. The relationships remain inspectable and source-linked rather than existing only in a separate graph database.
Derived access structure	A rebuildable index, embedding store, graph projection, summary, extracted-fact table, menu, or cache used to locate or present canonical sources, artifacts, and accepted ground.
Registration	The governed process that records a resource’s stable identity, type, version, owner, scope, provenance, standing, source path, relationships, and relevant validation records in canonical owner files, making it resolvable. Registration alone does not grant acceptance, activation, or authority.
Artifact or source admission	The person-authorized operation that gives an existing external or unmanaged file stable local identity, an object class, provenance, and a custody state. Admission makes the object addressable; it does not establish truth, acceptance, promotion, activation, or authority.
Registry	A rebuildable layer that compiles declared identities, scope, standing, relationships, and lifecycle state from canonical owner files for validation, lookup, and Atlas views. It does not create those declarations or their authority.
Flow Atlas	A generated visual and query layer showing definitions, ownership, lineage, consumers, impact, contradiction, staleness, and system flow. It remains subordinate to canonical files.
Receipt	Evidence that a bounded operation was authorized, attempted, applied, returned, accepted, or failed. It proves an event record, not the semantic quality of the result.
Reconstruction	Rebuilding the operative semantic situation after a restart, interruption, model change, or deletion of derived indexes.

Authority, judgment, and safe effects

Term	Meaning in AIOS
Bounded semantic sovereignty	A model’s jurisdiction to make genuine judgments of meaning over a composed situation without acquiring unrestricted operational authority.
Semantic jurisdiction	The purpose, object, ground, expected result, and consequence boundary inside which a model is asked to judge.
Semantic custody	The responsibility to carry meaning, scope, standing, provenance, and intended relationships accurately across people, models, artifacts, and operations without the machinery becoming the author or independently granting acceptance.
Deterministic custody	Exact control of stable identities, permissions, structural validation, target resolution, file effects, and records by non-semantic mechanisms.
Mechanical observation	Detection of declared state such as file changes, invalid syntax, stale hashes, version mismatch, or reconstruction need without interpreting human meaning, initiating semantic work, or granting authority.
Semantic initiation	Admission of a bounded cognitive movement that interprets a situation, produces a judgment, or proposes a consequence under declared ground and authority. Mechanical observation alone cannot supply it.
Mechanical invariant	A condition that must be enforced exactly for truthful or safe operation, such as current target identity or a destructive-action permission.
Semantic norm	A generally useful, scoped practice that guides judgment but remains open to exception and revision.
Governing decision	An authorized choice that directs a declared scope until it is reopened, revised, or superseded.
Authorized departure	A scoped, authority-bearing decision to depart from applicable guidance while preserving the guidance, circumstance, rationale, consequence, and result as reviewable evidence. A departure can motivate later review but does not automatically rewrite the guidance.
Recommendation	A contextual proposal for a next movement. It remains inert until selected or covered by an existing bounded grant.
Semantic action field	A visible, bounded set of meaningful next possibilities over which a model can recommend, a person can steer, and exact software can resolve consequences.
Trusted action path	The chain that binds a person’s approval to the canonical target, exact parameters, computed change, and operation that actually executes.
Five validity layers	Syntax, semantics, identity, authorization, and effects. Success at one layer does not establish success at the others.
Negative authority	Evidence that prohibited or ungranted effects did not occur, not merely evidence that an allowed result was produced.
Compliance-enabling	Providing records, controls, provenance, review, and bounded effects that can support a regulated implementation. It is not a declaration of legal compliance or certification.
Attestable domain package	A versioned domain knowledge system or reusable expertise package whose sources, scope, controls, evaluation, lineage, and integrity can be independently examined. “Certified” is reserved for an actual certification issued under a named authority.
Local knowledge-system fleet	A set of separately governed local installations that may receive a shared domain package. Every installation retains explicit version, compatibility, acceptance, variation, departure, revocation, and rollback state; publishing one version does not make every copy current or conforming.

Work, attention, and distributed intelligence

Term	Meaning in AIOS
Plan	A live, revisable theory connecting a desired outcome to strategy, artifact architecture, commitments, foreground work, and completion evidence.
Branch	A bounded independent line of work that remains related to a parent question, plan, receiving object, and return condition.
Commission	The complete contract for delegated work: purpose, charge, granted and withheld ground, scope, tools, authority, expected product, receiving object, and stop condition.
Return	An evidence-shaped delegated result awaiting sufficiency review and reintegration.
Return sufficiency	Whether the result answers the charge, exposes evidence and limitations, and makes its possible delta for the receiving object legible.
Attention-following asynchrony	An interaction architecture in which the person can move among objects while admitted work remains bound to its origin and returns without stealing focus.
Foreground movement	The human-sized act serving the person’s present attention.
Context horizons	Three related resolutions of contextual availability: the immediate frame of the present object and movement; the active semantic situation of the artifact, project, and live branches; and the durable domain ground of purposes, knowledge, decisions, relationships, practices, and lineage. They remain concurrently addressable at system level but are not injected wholesale into one model call.
Cross-resolution context lock	The degree to which a bounded context aligns the immediate frame, active semantic situation, and durable domain ground around purpose, identity, standing, and current state. It is movement-relative alignment, not a guarantee of completeness or correctness.
System-level concurrency	The capacity of the wider architecture to preserve multiple context horizons, branches, and cognitive movements at once through files, metadata, and separately bounded processes. It does not imply that one model invocation attends to the whole system simultaneously.
Concurrent bounded frame	One delegated branch’s limited world: its inherited purpose, immediate charge, warranted evidence and domain ground, authority, and return contract, preserved alongside other frames without silent merging.
Authorized asynchronous work	Deeper work that may continue under a bounded grant and explicit return conditions. It is not permission to invent adjacent objectives.
Cognitive surrender	A risk frame for the loss of active human engagement, understanding, judgment, or retained capability as systems perform more cognitive work. AIOS is designed to mitigate the risk; the term does not name a universally established syndrome.
Co-adaptive human–AI relationship	A continuing governed relationship in which a person's questions, practices, corrections, and accepted judgments change the durable semantic environment, while that renewed environment changes what can be brought into focus and made possible next. It is sociotechnical co-evolution, not reciprocal consciousness or equal authority.
Distributed cognition	Analysis of capability across coordinated people, models, artifacts, representations, and operations rather than within one component alone.
Branch budget	The discipline of comparing the value of independence or parallelism with coordination, duplication, collision, and reintegration cost.
Completion proof	Evidence that the intended outcome and criteria were met, including residual uncertainty, defects, deferred work, accepted risk, and reopen conditions.

Local intelligence and economic terms

Term	Meaning in AIOS
Local-first	Canonical files, accepted ground, and ordinary control remain in a person- or organization-controlled local environment, while explicitly authorized remote computation or sharing may still occur.
Local custody	The person or organization retains possession and control of durable files, context, judgments, workflows, and history.
Privacy sovereignty	The capacity to know and govern what data exists, where it moves, which model sees it, and under whose authority. Local storage contributes to it but does not guarantee it.
Model agnosticism	Durable intelligence is represented outside one provider so different models can use it, subject to adaptation and revalidation.
Dynamic model selection	Policy-gated selection of a model–harness configuration for each bounded movement according to task fit, privacy, consequence, quality, latency, cost, and verified history.
Provider portability	The ability to preserve durable ground and operating contracts while changing model providers, with explicit migration and revalidation costs.
Capability substitution	The hypothesis that stronger scaffolding can compensate for some raw model-capability difference on bounded, evidenced, and verifiable work. It is not general model equivalence.
Frontier complementarity	The use of the same reasoning architecture to give frontier models better ground and reserve their capability for movements where it materially changes the outcome.
Accepted-outcome economics	Cost measured across inference, latency, review, correction, reconstruction, governance, migration, energy, and failures per result that is actually accepted—not merely price per token or per response.

Relationship vocabulary

The knowledge system becomes navigable through explicit relationships. These verbs should remain stable across text, metadata, diagrams, and visual maps.

Relationship	Meaning
derives from	Was produced through a declared transformation of source ground.
cites	Uses as evidence or lineage without claiming derivation.
interprets	Provides a reasoned reading of a source.
implements	Realizes an architecture or contract in executable behavior.
validates	Supplies evidence against declared criteria.
consumes	Uses an object as input or governing ground.
produces	Creates a declared semantic or mechanical result.
projects	Presents a derived view without becoming a canonical source or artifact or independently acquiring accepted standing.
narrows	Retains a claim within a smaller valid scope.
qualifies	Adds a condition or limitation without replacing the complete prior claim.
revises	Changes content while preserving identity and lineage.
supersedes	Replaces current governing standing for a declared scope.
contradicts	Cannot jointly govern the same time, scope, and interpretation.
challenges	Supplies an objection or alternative that may require review without being a direct contradiction.
requires	Cannot operate or be accepted without the related object or condition.
recommends	Proposes a movement without authorizing it.
accepts	Grants standing within declared scope and authority.
registers	Assigns stable identity, scope, provenance, version, and relationships.
activates	Includes relevant accepted ground in a particular context.
retires	Removes an object from ordinary activation while preserving appropriate history.

Distinctions that should never collapse

Standing from evidence to reintegrated ground

How evidence may support interpretation and proposal, while acceptance, activation, effect, and reintegration remain separate transitions.

Semantic standing · canonical28-glossary-source-guide-and-research-traceability--m01.mmd

The repeated phrase 'may support' prevents storage or lineage from becoming truth. Each later edge adds a different requirement, so a produced result reaches an exact effect only after validation and authority, then becomes future ground only through review.

In particular:

provenance is not truth;
retrieval is not relevance;
a summary is not its source;
structural validity is not semantic correctness;
semantic correctness is not authorization;
authorization is not evidence that an outcome was good;
a recommendation is not an action;
a produced result is not accepted knowledge;
an agent role is not a new authority;
system-level concurrency is not simultaneous omniscience inside one model invocation;
a context horizon is not a semantic altitude or a cognitive operation;
co-adaptation is not reciprocal consciousness or equal authority;
a higher representational level is not a higher permission level;
internal consistency is not agreement;
local storage is not automatic privacy;
auditability is not regulatory compliance;
model agnosticism is not equal model capability;
and repeated practice is not automatically reusable expertise.

How research enters the publication

The product architecture is the governing narrative. External research has three roles:

Support: show that an underlying problem or mechanism is consequential.
Refine: make an AIOS distinction more precise or measurable.
Challenge: identify a condition, counterforce, or failure mode the architecture must address.

Only direct evaluation of AIOS can validate the integrated system.

The evidence chain has four resolutions:

Four resolutions of research evidence

How a public mechanism connects to a concise evidence synthesis, full analysis, and the original paper or authoritative report.

flowchart LR P["Public mechanism and claim"] --> O["Original paper or authoritative report"] P --> B["Normalized research brief"] B --> M["Full research memo"] M --> O

Source descent28-glossary-source-guide-and-research-traceability--m02.mmd

Text equivalent

Public mechanism and claim → Original paper or authoritative report; Public mechanism and claim → Normalized research brief; Normalized research brief → Full research memo; Full research memo → Original paper or authoritative report.

The public chapter explains the architecture first. A reader can then descend through the brief and memo, or move directly to the original source; both paths keep the external evidence attached to the exact statement it supports.

The public chapter explains the architecture first and links the strongest original sources near the claims they support.
The normalized brief explains the relevant findings, boundaries, integration value, and exact route into the full analysis.
The full memo preserves methods, source comparisons, counterevidence, and the larger bibliography.
The original paper or report remains the final external source for a research claim.

The Research Library index provides access to all eight research programs.

Source selection and freshness

AI research, model performance, market structure, regulation, cost, and hardware conditions change rapidly. The publication therefore uses a recent-evidence default:

current technical, capability, deployment, security, economic, and market claims should normally rely on research or authoritative material from the preceding 24 months;
current laws, regulations, standards, and policies should be checked against their operative text at publication time;
older sources should appear only when they provide indispensable conceptual lineage rather than evidence of current performance;
peer-reviewed primary research is preferred for empirical technical claims;
preprints are identified as preprints and treated as provisional;
vendor research can establish what the vendor reports about its own system but does not independently establish broad comparative conclusions;
volatile statistics and product capabilities should be rechecked immediately before publication.

Recent does not automatically mean reliable, and peer review does not remove the need to inspect methods. The relevant source is the one whose design actually supports the sentence being written.

The eight research programs

Research program	What it contributes	Entry point
Human cognition, agency, and cognitive surrender	Reliance, learning, transfer, meaningful human judgment, and agency measures	Normalized brief
Emergent and distributed cognition	Operational emergence, system unit of analysis, coordination, and analogy boundaries	Normalized brief
Relational memory and knowledge representation	Temporal state, plural retrieval, graph limits, provenance, and memory security	Normalized brief
Context composition	Effective context, information placement, retrieval alternatives, and evidence budgets	Normalized brief
AI-native architecture and bounded sovereignty	Semantic judgment, deterministic effects, trusted execution, and agent security	Normalized brief
System capability and model routing	Harness lift, interaction effects, capability substitution, routing, and evaluation	Normalized brief
Local and sovereign AI	On-device feasibility, privacy boundaries, portability, and regulated deployment	Normalized brief
Economics and institutions	Semantic capital, accepted-outcome economics, organizational adoption, and decentralization	Normalized brief

Maintaining the knowledge base

The publication should evolve through the same principles it describes:

Revise a concept at its primary page rather than introducing a competing definition elsewhere.
Identify every chapter, diagram, glossary entry, and research claim affected by the change.
Preserve material disagreement or prior form when understanding the change requires it.
Update links and source descent together with the prose.
Recheck Mermaid diagrams, examples, and journeys against the revised mechanism.
Keep research secondary to product architecture while allowing evidence to narrow, challenge, or reopen a hypothesis.
Test the knowledge base for orphan pages, redundant definitions, broken relationships, and inconsistent standing.

This approach allows the overview to remain a living, internally consistent knowledge system rather than a sequence of documents that slowly drift apart.