SEPT Principles — Basic Definition

Shared / Evolving / Plural / Traceable

Koji Annoura

1. Introduction — simplifying knowledge is not the same as losing meaning

Generative AI can analyze large volumes of information and generate responses. However, retaining every LLM response does not, by itself, constitute effective knowledge management.

Knowledge formation involves selecting, organizing, and connecting information while maintaining appropriate access to source data. The resulting knowledge should be understandable and verifiable by its intended human users and usable by AI systems.

A recurring pattern across several observations may become a rule or hypothesis. Complex decisions may depend on identifiable concepts, assumptions, conditions, and relationships. Explicit representation of these elements can support reuse in other contexts.

However, abstraction, generalization, and integration may remove significant context, exceptions, or alternative interpretations. A clear explanation is not necessarily a faithful account of how a decision was made.

Simplifying knowledge is not the same as losing its meaning.

An abstraction should retain sufficient links to its assumptions, alternative interpretations, sources, and supporting evidence to permit subsequent examination.

Knowledge is not a finished product

Physics identifies patterns through observation and tests their scope of applicability. Mathematics generalizes structures and clarifies definitions and conditions. These provide analogies for knowledge formation; however, empirical hypotheses and mathematical proofs remain fundamentally distinct.

When new evidence challenges an existing rule, its conditions and associated hypotheses require reassessment and, where appropriate, revision and further validation. Simply adding a condition does not make a claim true.

Knowledge should remain open to revised conditions, corrections, and further generalization in response to new evidence.

SEPT establishes design considerations applicable from the creation of knowledge, not solely after its storage.

Origins of SEPT

SEPT emerged from the author’s long-term practical experience in collecting, validating, transforming, integrating, and modeling data.

Long-running business systems accumulate temporary fixes, direct data patches, and changes in business rules. Consequently, some records and inconsistencies can no longer be explained by the current specifications or data model.

Migration to a new system is not necessarily a one-time event. Across successive migrations, exceptional or inconsistent records may be carried forward, while the meaning of original values and the reasons for earlier corrections become difficult to reconstruct.

Integrating different data sources presents related challenges: fields with similar names may have different definitions or classification schemes. Work with Japanese COVID-19 open data exposed differences in age categories across prefectures, as well as errors in primary records.

These experiences motivated the need to retain not only current data values, but also their meanings, origins, change histories, and relationships to alternative definitions.

SEPT was formulated from these practical challenges, rather than derived from a particular existing theory or framework. Related research and technologies address overlapping concerns, but should not be confused with the original sources of the proposal.

2. Basic definition of SEPT

2.1 The four principles

SEPT consists of four design principles to consider when creating, abstracting, generalizing, integrating, updating, sharing, and reusing knowledge, so that its meaning and supporting evidence are not lost. The same principles can also serve as criteria for evaluating knowledge and knowledge systems.

SEPT has two roles: it guides decisions during design and knowledge formation, and it provides perspectives for evaluating the resulting knowledge and systems.

It prescribes no specific AI model, database, communication protocol, data model, or business framework. It does not guarantee the truth of knowledge or supply an algorithm for finding the best generalization.

2.2 Scope of knowledge addressed

SEPT does not claim to settle philosophical definitions of knowledge. Its practical subject includes claims, interpretations, hypotheses, rules, and decisions derived from observations and records. These should not be treated as interchangeable.

A record stating that A occurred does not, by itself, establish that A actually occurred. Likewise, observing that method Y worked in one situation differs from asserting that Y generally works under the same conditions. The generalized claim requires additional validation and a defined scope of applicability.

2.3 Knowledge Sustainability

Within SEPT, Knowledge Sustainability means the capacity to understand, verify, and reuse relevant knowledge—including its meaning, conditions, alternative interpretations, evidence, and changes—even as people, organizations, and technologies change over time.

This is more than retaining data. A design document may remain available while its original intent, constraints, and rationale are no longer recoverable. SEPT seeks to support sustainable knowledge by applying the four principles early in its lifecycle. The principles alone, however, cannot guarantee sustainability.

3. The four SEPT principles

3.1 Shared — understandable and transferable knowledge

Shared means keeping knowledge from being confined to a single person or system so that its intended users can understand its meaning and context, carry it forward, and reuse it.

The transfer of information does not necessarily constitute the sharing of knowledge. An experienced engineer may make good decisions from tacit experience, yet others may be unable to repeat the reasoning. Even where design documentation remains available, the rationale for specific decisions may be unavailable to subsequent engineers.

Documents, diagrams, models, training, demonstrations, and collaborative work can all convey context and reasoning.

Human-understandable generalization

AI systems can process complex information; however, outputs accessible only within a particular system may be difficult for human users to interpret and transfer. Identifying the important concepts, conditions, and relationships behind a decision may make knowledge easier to transfer.

Not every complex decision can be reduced to a simple rule, and a fluent explanation may not faithfully describe the original decision process. Simplification must not compromise material conditions or evidence.

The relevant criterion is whether the intended users can understand the material meaning and assumptions, and verify or reuse the knowledge. Human comprehension and machine processing are different capabilities and may require different representations. Shared remains subject to confidentiality and access-control requirements; it does not require unrestricted disclosure.

Design consideration: Identify intended users, purposes of use, and the representations needed for comprehension and reuse.

3.2 Evolving — preserving change and development

Evolving means distinguishing changes in the world, corrections to records, and changes in knowledge or interpretation—and making those developments understandable.

When an interpretation is revised following the discovery of a historical source, the underlying past event remains unchanged; its interpretation has changed.

Reviewing Japanese prefectural COVID-19 open data also revealed errors in primary records. Some errors were reported to the relevant prefectures for correction. These were corrections to published records, not changes to the underlying events. A knowledge system should distinguish the earlier record, the corrected record, and the reason for correction.

A hypothesis that method X is effective under condition A may encounter a counterexample indicating that condition B is also relevant. Revising the claim to “A and B” still requires testing. The prior hypothesis should remain distinguishable, together with the evidence supporting it and the reasons for its revision.

Evolving requires no particular review cadence or workflow.

Design consideration: Accommodate subsequent additions, corrections, reassessment, and further generalization.

3.3 Plural — coexistence of distinct meanings and interpretations

Plural means retaining different records, classifications, observations, claims, perspectives, and interpretations about the same subject or event, together with their meanings and contexts, without requiring them to merge into one conclusion.

Plural is particularly important to this proposal. Different records may not mean the same thing. Even the same evidence may support more than one interpretation.

Different classifications and granularity

A practical motivation for this principle arose from integrating Japanese COVID-19 open data. Prefectures used different fields, definitions, and age classifications.

“Under 10” is an age-based classification, whereas “not yet enrolled in school” is based on school enrollment status. They are not equivalent categories.

Similarly, “90 and over” includes people aged 100 or older, whereas separate “90–99” and “100 and over” groups preserve finer detail. Fine-grained categories can be aggregated when their definitions are compatible, but a broad category cannot be split accurately without additional evidence.

When data is transformed into a common representation, retaining original categories and transformation rules can prevent loss of meaning.

Different interpretations

Historical researchers can interpret the same documents differently because of their questions, methods, or priorities. One may emphasize the economic benefits of a reform, while another examines its social costs. These perspectives are not necessarily direct contradictions.

A business team may also disagree about investment despite using the same market data. Different forecasts, risk tolerance, objectives, and values can lead to different decisions. Rejected options may remain useful when circumstances change.

Preserve differences without suspending judgment

Distinct information or interpretations should not be merged without establishing their equivalence. Keep the source claim, its meaning, conditions, context, relationships, and unresolved differences where needed.

Plural does not imply that every claim is equally correct. Incorrect records should be corrected, and unsupported hypotheses should be distinguished from supported ones. An evaluation or decision should not inadvertently erase the existence or context of alternative records or interpretations.

Design consideration: Preserve meaningful differences and support comparison and evaluation without requiring unjustified consolidation.

3.4 Traceable — provenance and use of knowledge

Traceable means maintaining explicit relationships between knowledge and its sources, evidence, formation and transformation processes, decisions, and downstream uses so that these relationships can be examined.

A statistic is difficult to assess without its source records, reporting period, and calculation method. Connections to primary records and transformation steps make it possible to verify how a result was derived.

The same applies to generalized claims. Identifying the observations supporting a rule enables targeted reassessment when counterexamples emerge.

Business decisions can be represented through a sequence such as:

Observation → Hypothesis → Decision → Action → Outcome → Evaluation

A successful outcome does not, by itself, establish the quality of the decision process. Preserving the assumptions, alternatives, and rationale helps people learn without blindly repeating past choices.

Traceability does not establish the truth of a claim.

Design consideration: Maintain appropriate links among knowledge, sources, transformations, rationale, and subsequent uses to support verification and revision.

4. Applying SEPT across the knowledge lifecycle

SEPT guides design decisions from creation through reuse rather than merely assessing a completed system.

These are not mandatory sequential phases. Activities may overlap or repeat.

Start with the knowledge needed now

A complete ontology or model is not a prerequisite for initial development. One iterative approach begins with the knowledge required to answer a current question, validates that knowledge through application, and extends it as new questions arise:

Question → Minimum Knowledge → Apply & Validate → Identify Gaps → Extend & Generalize

This iterative approach is compatible with SEPT, but SEPT does not mandate this exact process.

Minimum Knowledge is not Minimum Data

Constructing only the knowledge needed for a current question does not justify discarding its supporting source data. For example, when deriving common failure conditions from incident reports, the original reports should remain available so that circumstances and exceptions can be examined. If later evidence identifies another relevant condition, the hypothesis can be reassessed against those records.

How the principles work together

Preserving different meanings or interpretations involves Plural, while examining their supporting evidence involves Traceable. Understanding how knowledge has changed involves Evolving. Even when every source is identified, Shared may remain weak if the intended reader cannot understand the meaning.

The four principles are related but provide distinguishable design perspectives.

5. Relationship to established work and technology

SEPT does not claim originality for each underlying concept. FAIR Principles address reuse; W3C PROV covers provenance; version control and temporal databases address change; semantic heterogeneity addresses differences in meaning; and knowledge representation addresses claims and relationships.

SEPT brings meaning, differing interpretations, change, and evidence together as design concerns throughout creating, integrating, updating, sharing, and reusing knowledge.

Graph data models

SEPT is technology-independent. Relational databases, document systems, and other architectures may support it. Graph data models are one option for representing relationships among claims, evidence, interpretations, temporal information, and decisions.

Property Graph and RDF technologies are possible implementation options. The adoption of a graph database does not, by itself, establish conformity with SEPT; appropriate data modeling remains necessary.

Knowledge Versioning

Knowledge Versioning is a concept for managing changes to knowledge, including history, differences, alternative interpretations, and revisions. It draws on version-control concepts associated with Git, but does not require the use of Git and does not denote a Git-based product. Unlike software changes, distinct interpretations do not necessarily constitute conflicts and need not be merged. Changes to meaning, assumptions, and evidence also differ from line-based text changes.

Knowledge Versioning and Knowledge Graph Diff are potential implementation approaches, not mandatory components of SEPT.

6. Evaluating knowledge and knowledge systems

SEPT can inform assessments of knowledge representations, their formation and transformation processes, and the systems that manage them. A Knowledge System here can include people, organizations, processes, documents, data, and software—not only a computer application.

An evaluation should specify its knowledge domain, intended users, goals, timeframe, and system boundaries.

These questions complement, rather than replace, assessments of accuracy, completeness, and usefulness.

Example: correcting source records and assessing impact

Consider a correction to an erroneous primary record, as occurred during the review of COVID-19 open data. If the original record has already contributed to aggregates or analyses, silently overwriting it makes it harder to reconstruct earlier results or identify downstream effects.

This case primarily concerns Evolving and Traceable: can the system distinguish records before and after correction and explain the reasons for change (Evolving), and can it identify affected aggregates or decisions (Traceable)? Shared is also relevant when users need to understand the correction. Evaluation should concern the implemented design, rather than the use of the term “SEPT.”

These are initial evaluation questions, not a formal certification standard.

7. Scope and limitations

SEPT may be applied to human knowledge transfer, science, history, business decisions, data integration, software design, and AI systems. It does not define or guarantee:

• An algorithm for extracting or optimally generalizing knowledge.
• Truth, completeness, or automated validation of every claim.
• A method for building a complete ontology or data model.
• Development frameworks such as Scrum, or specific business processes.
• A particular AI model or database.
• Permanent retention of all data.
• Knowledge Sustainability solely from satisfying four principles.

Not every complex process can be reduced to a simple rule. Security, privacy, lawful deletion, data integrity, preservation, and operational practices also matter.

It has not been established that the four principles are necessary or sufficient for Knowledge Sustainability. Their value and limits require evaluation through practical cases and implementations.

Conclusion — consider SEPT from the start

Knowledge development may begin with the requirements of a current question, rather than a complete knowledge model, and proceed through extension and revision in response to new evidence. Throughout this process, relevant meanings, alternative interpretations, change histories, and evidence should remain accessible.

Knowledge derived through AI-assisted processes should not remain inaccessible outside a particular model. Relevant concepts, assumptions, and relationships should be represented in forms that intended human users can understand and verify and that AI systems can process.

SEPT provides four design perspectives for deciding what to preserve, distinguish, and make verifiable throughout the knowledge lifecycle. The same principles can also support evaluation of the resulting knowledge and knowledge systems.

Related Research and References

1. Koji Annoura, Where Should Organizational Knowledge Live? Building AI Systems Beyond LLM Memory, Synapt AI.
2. Wilkinson et al., The FAIR Guiding Principles for Scientific Data Management and Stewardship, Scientific Data, 2016.
3. W3C, PROV-O: The PROV Ontology.
4. W3C, Time Ontology in OWL.
5. Schwaber and Sutherland, The Scrum Guide.
6. Git Documentation.

SEPT (Shared, Evolving, Plural, Traceable) is a set of knowledge design principles proposed by Koji Annoura. This working draft outlines its concepts, scope, and preliminary evaluation perspectives. It is neither a technical standard nor a conformity specification.