An education system cannot become more intelligent merely by collecting more links, papers, lessons and AI-generated explanations. It must know where each claim came from, what it means, whom it applies to, whether the material can lawfully be used, and what should change when the source changes. A curriculum reference that was authoritative last year may be superseded. A helpful diagram may have a licence that does not permit commercial reuse. A published study may later receive a correction. A confident AI answer may cite the wrong edition. Without a reliable reference registry, these errors can spread from one resource into thousands of lessons.
Ministry of Education V3.0 therefore needs a knowledge-provenance protocol: a stable, searchable record of official syllabuses, concepts, evidence, teaching resources, permissions, relationships, reviews and corrections. This reference explains how to run such a registry in schools and across a national education system. It includes a ten-stage intake procedure, source authority ladder, rights gate, canonical concept record, HELP and REF separation, drift control, example case studies, a classroom task and a working quality rubric.
This is an independent eduKateSG educational architecture, not an official Singapore Ministry of Education data infrastructure or a declaration that external providers have endorsed it. Source quality, usage rights and legal authority must be verified against the current original document. The framework helps people use knowledge responsibly; it does not remove professional responsibility to teach, check and correct.
1. A Library Is Not the Same as a Reliable Knowledge System
Imagine a secondary Mathematics teacher preparing a lesson on solving equations with fractions. Search results return a commercial worksheet, an old examination document, a university lecture, a parent blog and an AI explanation. Each may contain useful ideas. But only some may align with the relevant student’s level and syllabus. Some may contain material that cannot be republished; some may assume prerequisite knowledge the class lacks.
A dependable knowledge system does not place all five results into a folder titled ‘fractions’ and treat them as equivalent. It gives the underlying concept a stable identity, shows prerequisite relationships, distinguishes official requirements from complementary resources, records authority and update dates, and checks how the specific item may be shared or taught.
The distinction becomes more important with AI. A generated response can combine multiple sources into one attractive explanation, but the combination may conceal disagreement, outdated statements and incompatible permissions. Provenance must survive summarisation. What a learner sees should be an intelligible explanation connected to checked references, not an untraceable assertion.
2. The Eight Rooms of a Knowledge Crosswalk
| Room | Essential function | Minimum control |
|---|---|---|
| 1. World source registry | Catalogue source name, owner, type, official status and origin | Verified URL or identifier and date |
| 2. Canonical concept library | Give each teachable concept a stable meaning and ID | Definition, scope and controlled aliases |
| 3. Crosswalk engine | Describe prerequisite, broader, narrower, related and confusion relations | Typed, reviewable links rather than arbitrary associations |
| 4. Curriculum mapper | Align resources with the appropriate stage, subject and version | Official source and validity period |
| 5. REF evidence store | Track claims, authorship, studies, corrections and confidence | Traceable evidence and retraction check |
| 6. HELP teaching store | Provide age-appropriate examples, checks and teaching sequences | Learner level, misconceptions and permissions |
| 7. Runtime guidance | Select an explanation for the learner and context | Age, licence, curriculum, evidence and safety gates |
| 8. Return and drift control | Observe changed sources and failed explanations | Review owner, corrected item and revision record |
These rooms work together but should remain conceptually distinct. A source registry answers ‘Where did this come from?’ A concept library answers ‘What does it mean?’ A curriculum mapper answers ‘When and where is it required?’ A teaching packet answers ‘How should it be explained?’ An evidence store answers ‘Why should this claim be trusted?’ Blurring those questions invites avoidable mistakes.
3. Separate Official Authority, Evidence Strength and Teaching Usefulness
Three qualities are often mistaken for one. A document may be authoritative because it defines an examination requirement. A research study may provide strong evidence for a teaching method without having any authority to define a national syllabus. A carefully designed classroom example may be pedagogically useful without itself being a source of policy or scientific proof.
| Question | What determines it? | Example |
|---|---|---|
| Is this an authoritative requirement? | Issuer’s legal or institutional mandate and valid version | Current examination specification from the responsible board |
| Is the claim well supported? | Relevant methods, quality, scope, replication and corrections | Appropriate synthesis of educational research |
| Is the resource good for this learner? | Accuracy, level, accessibility, feedback and prerequisites | Teacher-created worked example |
| Can it be reused? | Copyright, licence, permissions and applicable restrictions | Openly licensed educational material with verified terms |
| Can it be updated? | Owner, identifier, source location and change-management route | Versioned concept and resource record |
The purpose is not to rank every external organisation on one simplistic scale. For questions about Singapore assessment requirements, a current SEAB publication has a different kind of authority from a scholarly article. For questions about scientific findings, evidence must be assessed by method and relevance. For classroom use, the teacher must also verify suitability for the child.
4. The Ten-Stage Source Intake Protocol
Step 1 — Define the educational need
Record the question the source is intended to answer. ‘Secondary 1 learners need to connect numerical fractions and algebraic expressions’ is a specific instructional need. ‘We want more AI knowledge’ is too broad for meaningful validation. The destination determines which kind of authority and evidence is required.
Step 2 — Identify the original source and owner
Find the original publication, issuing body, author or editor where identifiable, document title, accessible URL and stable identifier when available. Record whether the content is a primary source, secondary synthesis, teaching illustration or platform-generated summary. A copied PDF is not necessarily the authoritative edition.
Step 3 — Check the exact version
Store issue date, effective date where relevant, retrieval date and revision number. Preserve the earlier citation as historical evidence when the rules permit it, but mark that version as superseded for current use. A curriculum system should not silently replace a source while continuing to claim it is the same edition.
Step 4 — Verify factual and evidential scope
Identify the exact claim being supported, its limitations, relevant population and confidence. A study conducted with older students in a particular context cannot automatically establish outcomes for all primary learners. Record contradictory evidence and corrections rather than ignoring them.
Step 5 — Run the rights and licensing gate
Distinguish permission to access or link to material from permission to copy, transform, redistribute, commercially exploit or ingest into an AI system. Verify the source-specific licence and any additional restrictions, and preserve required attribution. If the rights are uncertain, default to a properly cited link rather than copying the material into a public teaching packet.
Step 6 — Map the canonical concept
Connect the source to a clearly defined concept, including its scope and aliases. For example, ‘equivalent fractions’ and ‘fraction equivalence’ may refer to the same underlying idea; ‘simplifying algebraic fractions’ is a related but more advanced application. A stable identity helps a teacher find resources without turning differently named topics into disconnected silos.
Step 7 — Record typed relations
Use relationships such as exact match, close match, broader, narrower, prerequisite, assessed by, confused with, corrected by and transfer to. Do not label a weak relationship as exact alignment because an automated similarity score is high. Human review is essential for consequential curricular mappings.
Step 8 — Build a learner-facing HELP packet
Develop an explanation at the relevant stage, one worked example, a predictable misconception, a short practice sequence and an independent check. Include accessibility considerations and safe alternatives to digital-only materials. The packet may link to external resources without republishing content whose rights do not permit that use.
Step 9 — Issue an REF receipt
Attach a compact evidence note identifying the supporting sources, authority class, date, applicability, important uncertainty, rights status and reviewer. Mark whether an AI-generated statement has been independently checked. Do not treat AI fluency or a model’s confidence as a substitute for provenance.
Step 10 — Assign maintenance and drift triggers
Set a responsible owner and review date. Trigger a new check when a syllabus changes, a link breaks, a journal paper is corrected or retracted, a provider modifies licensing, a tool changes behaviour or learners repeatedly misunderstand an explanation. A reference without maintenance becomes a future liability.
5. A Minimum Viable Source Registry Card
| Field | Required entry | Example |
|---|---|---|
| Source ID | Stable registry identifier | SOURCE-SG-MATH-0001 |
| Original source | Owner, title, URL and optional DOI | Applicable publisher and reference |
| Source class | Official syllabus, study, repository, teaching material or context | Official curriculum source |
| Authority and scope | What the source is empowered or suitable to establish | Subject outcomes for specified stage |
| Version | Issue, effective, retrieval and last review dates | 2026 review record |
| Licence and usage | Link, copy, adapt, AI input, commercial and attribution rules | Link permitted; copying unverified |
| Canonical concept | Stable ID and reviewed relationship | MATH-FRACTION-EQUIVALENCE |
| Learner level | Relevant stage and prerequisites | Lower secondary bridge after primary fractions |
| Evidence quality | Method, limitation and correction status | Official standard; not a causal study |
| Owner and expiry | Reviewer and next review trigger | Curriculum editor; check annually or on syllabus change |
A registry can begin as a carefully maintained table. It does not require complex software on the first day. The essential discipline is that identifiers, sources, versions, permissions and relationships remain legible when the system grows.
6. How Knowledge Standards Help Without Replacing Human Judgement
The 1EdTech Competencies and Academic Standards Exchange (CASE) standard provides a structured way for systems to exchange learning standards, competencies, alignments and associated rubrics. CASE 1.1 was issued in January 2025; it includes globally unique identifiers for standards when implemented within that specification. Schools may use compatible approaches when they need interoperable curriculum records.
The W3C Simple Knowledge Organization System (SKOS) supports concept schemes and semantic relationships, including broader, narrower and related concepts. SKOS can help catalogue and connect educational concepts but does not determine that a particular link is pedagogically valid or that a student’s prerequisite has been mastered.
A good knowledge system uses standards to prevent confusion between tools, not to shift curriculum authority from qualified educators to software. A technically valid machine-readable link can still be educationally misleading; evidence and human review must accompany interoperability.
7. Worked Example: An Algebra Bridge After Primary School
Illustrative editorial workflow. A teacher notices that students can add simple numerical fractions but fail on expressions such as x/2 + x/4. The concept library identifies three related nodes: fraction equivalence, like-denominator addition and symbolic representation of equivalent expressions.
The editor checks the current relevant syllabus, selects permitted examples and writes a HELP packet beginning with numerical fractions, moving into a visual representation and finally substituting algebraic terms. The REF packet distinguishes the official curriculum linkage from any evidence claims about the teaching method. It documents which specific classroom examples were independently written and which sources are linked but not reproduced.
A later independent check presents a different expression. If students still fail, the item receives a teaching-drift note: perhaps the explanation needs a better translation bridge. The source registry does not call the original mathematics incorrect; it records that the teaching resource is insufficient for the observed learning need. This is why knowledge truth and pedagogical usefulness must be assessed separately.
The Secondary 1 Mathematics tutoring example in Clementi illustrates the practical setting in which a precise starting-position diagnosis can lead to targeted instruction. It does not determine an official curriculum requirement.
8. Worked Example: A Public Claim Loses Its Original Conditions
A public article summarises research about a learning intervention and claims that it ‘raises academic success’. The original research concerns a narrower outcome in a specific student population. The REF editor links the original source, records the relevant study design and limits the conclusion. The HELP editor develops a class exercise asking students to identify where the generalisation occurred.
If a subsequent correction alters the finding, the registry can identify the dependent claims and classroom packets. An editor reviews them, changes the confidence statement where justified and records what was published before and after the correction. This is stronger than simply replacing a URL and hoping teachers notice.
9. The Copyright and AI-Ingestion Gate
| Proposed use | Question to verify | Safe default when uncertain |
|---|---|---|
| Link to external material | Is the link publicly accessible and suitable for learners? | Link with proper context and citation |
| Copy a passage or illustration | Does permission, exception or licence allow this use? | Do not reproduce; create original explanation |
| Adapt a teaching resource | Are derivatives and redistribution permitted? | Use a different appropriately licensed source |
| Use in a commercial product | Are commercial rights included? | Treat rights as unverified until confirmed |
| Submit to an AI service | Does the licence and applicable data arrangement allow ingestion? | Do not upload; use approved alternatives |
| Store student work | Are collection, retention and access authorised? | Minimise and keep in approved systems |
Not every ‘free’ resource is open for commercial adaptation or AI ingestion. Rights should be checked at the item level where necessary, including media embedded inside otherwise openly licensed pages. Data protection for student submissions is a distinct obligation from copyright permission for reference content.
10. A School Reference Library’s Monthly Quality Review
- Check a sample of the most used syllabus and assessment links for current editions.
- Inspect resources whose licences or publishers have changed.
- Look for unsupported causal claims and AI-generated citations that were never checked.
- Identify concept duplicates and falsely labelled exact matches.
- Review learner-reported confusing explanations and the related independent task results.
- Check published source corrections and retraction notices before repeating research claims.
- Examine whether age, language, accessibility and stage information remains accurate.
- Record repaired items, their owner and the next review trigger.
A monthly review can be proportional to system size; not every source requires identical checking frequency. High-stakes, fast-changing or widely reused items merit more frequent scrutiny than stable mathematical definitions. Risk-based scheduling saves effort while protecting learners.
11. A Thirty-Five-Minute Teaching Packet Audit
| Minutes | Learner or teacher task | Output |
|---|---|---|
| 0–5 | Choose an educational claim and define the actual learning need | One-sentence purpose |
| 5–12 | Trace original publisher, date and statement | Source identity record |
| 12–19 | Check what the source supports and which audience it addresses | Bounded evidence note |
| 19–25 | Identify concept alignment and copyright permissions | Reviewed mapping and rights status |
| 25–31 | Draft an age-appropriate explanation and fresh check question | HELP packet outline |
| 31–35 | State who will review the item and what change triggers a new check | REF receipt and review date |
The teacher should avoid making young learners responsible for checking legal permissions. Students can practise distinguishing source, explanation and evidence with preselected safe material. Licensing and official curricular sign-off remain institutional responsibilities.
12. Performance Standard for Registry Editors
| Criterion | Developing | Proficient | Advanced |
|---|---|---|---|
| Provenance | Retains a useful link only | Records original source, owner and dates | Can explain versions and dependencies |
| Authority | Treats every source as equivalent | Separates policy authority from research evidence | Identifies disagreements and source-specific limits |
| Concept mapping | Relies on keyword similarity | Uses reviewed typed relationships | Detects false matches and prerequisite confusion |
| Rights | Assumes publicly visible means reusable | Checks permissions before copying or adapting | Maintains granular rights and renewal controls |
| Return and repair | Leaves errors untracked | Corrects resources and records a revision | Finds dependent outputs and tests the correction |
13. The Gold Standard
A gold-standard education knowledge system is not the one with the largest warehouse. It is the one where a teacher can find the correct concept, the applicable current authority, trustworthy evidence, an appropriately licensed resource and a suitable teaching sequence—and can discover when any of those elements has changed.
The deeper objective is continuity. Knowledge can pass between publishers, syllabuses, schools, platforms, teachers and learners without losing its identity, authority or capacity for correction. A reference library becomes part of the civilisation’s learning infrastructure only when it knows how to preserve truth and repair error over time.
Connected Reference Library and Technical Foundations
- Ministry of Education V3.0 — Complete Reference Library
- Reference 07 — Curriculum Crosswalks and Learning Progression
- Reference 03 — Evidence and Claims Verification
- Reference 17 — Education Observatory and Drift Control
- Reference 12 — AI and Digital Learning Approval Gate
- 1EdTech — CASE Academic Standards Exchange
- W3C — SKOS Reference
- SEAB — Singapore Examinations and Assessment Board
- Singapore MOE — Syllabuses