How do you translate a Common Education Data Standards (CEDS) data dictionary so that families, teachers, analysts and international partners understand the same educational facts? CEDS Version 13 provides standard property definitions, relationships and option sets across learning and workforce systems. The professional task is not simply to replace English words in a column. It is to preserve a property’s identity, declared meaning, permitted values and reporting purpose while making its explanation readable in another language. A translation that looks polished but changes the definition of enrolment, attendance, programme participation or credential completion can quietly ruin an otherwise well-designed dataset.
This longform guide explains CEDS Version 13 localization, multilingual education data dictionaries, CEDS property definitions, Align maps, option-set translation, P–20W data integration, crosswalks, semantic equivalence, data governance and multilingual reporting. It is written for educational researchers, school and university data teams, ministries, data stewards, software developers, translators and accessibility specialists. It focuses on the language layer that humans read, the controlled metadata that software depends upon, and the tests that prove the two still describe the same evidence. A school does not need to adopt a particular country’s terminology in order to understand the distinction between a useful display label and a machine-readable standard.
The CEDS website provides downloads for Version 13 alongside previous versions; its tools describe properties, option sets, domains, entities and the Align workflow. CEDS is a common vocabulary and family of data models, not a magical universal translation service and not an assertion that education systems have identical legal categories. A defensible multilingual implementation records the source definition, identifies the local concept, marks unavoidable differences and preserves traceability. The central principle is simple: translate explanatory language and lawful presentation, but never silently redefine the original data contract.
Start here: the CEDS localization map
This article follows the core concepts, real-world decisions, a four-week learning plan and frequently asked questions. The examples are fictional and contain no student records. The goal is to preserve semantic identity while making education information useful across languages.
The essential foundations
What CEDS standardizes—and what it does not
A property in a data dictionary has more than a name. It may possess a unique identifier, a formal definition, a permitted format, an option set, an entity relationship and a purpose in reporting. A translated label provides an entry point, but it is not proof that two jurisdictions count the same people, institutions or outcomes. The first question for any localization project is therefore not ‘what is the closest word?’ but ‘which real-world condition causes this field to be populated?’ CEDS covers early learning, K–12, postsecondary, adult learning, careers and related domains; those areas intersect without collapsing into one school system.
Property identity is not a heading
Consider a data extract with a stable property identifier and a display heading reading ‘enrolment status’. A French, Mandarin or Arabic interface can render an appropriate local label without generating another property identity. Changing the underlying identifier to accommodate a translation breaks connections to source systems, dashboards and longitudinal comparison. Good multilingual data dictionaries retain exact technical identifiers in a protected field, publish localized labels separately and document which version of the English definition they explain.
Option-set values are categories, not sentences
An option set can contain allowable values with meanings that matter for machine interpretation. A human-friendly text string such as ‘unknown’ may mean not collected, not reported, withheld, not applicable or unavailable; these are not interchangeable. Translation should maintain the code and carefully distinguish its description from neighboring categories. When a local administrative framework has no one-to-one match, document that limitation rather than assigning a deceptively familiar word.
Why P–20W requires contextual humility
P–20W refers broadly to education and workforce pathways across stages of learning. Yet an age, grade, award or programme name need not mean the same thing internationally. A student can move institutions without having the same eligibility classification in the receiving system. A dictionary helps systems compare underlying facts; it cannot determine policy equivalence without policy context. Translate the definition and its limits, identify local rules separately and never turn a data mapping into an unfounded recognition of qualifications.
A sensible translation ledger
Build a working ledger containing source property ID, source version, exact source definition, source option values, target locale, proposed display label, translator’s rationale, local policy note, reviewer decision, date and a test reference. Store the original as a non-editable baseline. Reviewers must be able to recover why a local word was selected. A ledger makes corrections possible after new statutory terminology, revised definitions or changes to a school information system.
A crosswalk is a claim that needs evidence
A crosswalk says how one system’s fields relate to another’s. Sometimes the relationship is exact, sometimes broader or narrower, sometimes computed and sometimes unknown. Distinguish direct equivalence from approximate conceptual mapping, transformation and no mapping. The translated description must communicate these qualifiers plainly. If a district’s field describes a student registered for classes and an external system describes attendance on a census date, the two should not be merged because the English labels sound related.
Localization, accessibility and usability
Many readers meet education data in an export table or a small dashboard. Provide localized table headings, readable abbreviations, screen-reader labels and examples for technical categories. Do not hide an identifier that someone needs for audit. Define short acronyms on first use, and give a glossary route to the complete definition. Inclusive terminology should improve clarity without making assumptions about family structures, learning support, identity or achievement.
Where this article sits in the translation system
This guide concerns semantic data dictionaries and CEDS mapping, not course content packaging or learning event logs. For credential assertions consult the CLR 2.0 and Open Badges 3.0 guides; for grades and class rosters consult OneRoster 1.2; for learning activity statements consult xAPI 2.0. Each standard operates at a different layer. A multilingual education architecture becomes reliable when readers can travel between those layers without treating a convenient label as a universal definition.
Twenty-eight professional localization decisions
1. A pupil’s preferred name
A school has a legal administrative name, a preferred classroom name and an export identifier. One system puts all three under the loose heading ‘student name’. A receiving interface must show a respectful local display without confusing identification or substituting a preferred form into a legally required record. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Keep the identity reference and the distinct property definitions separate. Translate ‘preferred name’ in relation to its intended display purpose, rather than as a synonym for legal name or an instruction to rename the pupil. A school register is corrected in the wrong place, and an ordinary display adaptation becomes an unauthorized record change. Matching on a translated name also creates false duplicates when scripts or diacritics differ.
Worked acceptance test. Compare two synthetic pupils with similar names, a name change, and one learner whose display language differs from the institution’s language. Prove the exact source identifier survives. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
2. Enrollment versus daily attendance
A district exports annual registration and daily attendance to a national reporting system. The two variables are displayed side by side but one translator uses a single target-language word meaning ‘presence’ for both. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Describe enrolment as the status defined by the institution’s registration process and attendance as the specified participation or presence measure. Explain the reporting period in the label or help text where needed. Families think a registered child has been marked physically present, while analysts overstate participation by calculating an attendance percentage from enrolment counts.
Worked acceptance test. Use a child enrolled for a full week but absent for two days. The localized table must maintain separate rows, separate numerators and separate denominator definitions. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
3. School-year boundaries
One jurisdiction begins an academic year in January while a partner’s calendar begins in August. An export field called ‘school year’ is rendered without its calendar reference, and dates appear inconsistent. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Retain the declared academic-year identifier, start and end boundaries and time zone or local date convention. A locale’s customary school calendar is background context, not a replacement for source data. An August registration could be misclassified into the following year after an automatic locale conversion, corrupting enrolment and completion cohorts.
Worked acceptance test. Test dates at both academic-year boundaries, including a student transferring between jurisdictions. Review whether the report clearly states the source calendar. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
4. Grade level and age
A learner’s grade code resembles a numeric age in a translated interface. A reader assumes ‘Grade 7’ means every learner is seven years old, which is false. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Keep educational level or year-of-schooling distinct from chronological age. Explain the system’s naming convention and avoid making an unsupported equivalence to a different country’s class level. An eligibility service refuses access on the mistaken belief that a translated grade label is an age category.
Worked acceptance test. Show school-level, grade-level and birth-year information as separate concepts. Validate with a pupil who repeated a year and one who advanced early. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
5. Course versus course section
A university catalog holds one course definition and several scheduled sections with instructors, rooms and term dates. An import flattens both as ‘the course’. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Translate the catalog subject and section heading independently. The course is a curricular definition; the section is a particular offering with time-bound relationships. An advisor confuses course completion with enrollment in one scheduled section and cannot explain a timetable conflict.
Worked acceptance test. Use two sections of the same course, one cancelled and one active. Confirm the localized output retains the correct section identifiers. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
6. Programme participation
A learner receives language support, mentoring and an enrichment activity. The local expression for ‘programme’ sounds like a formal qualification. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Explain participation, service provision and credential award separately. A programme record describes an opportunity or service unless the authoritative data specifically records a completed qualification. A support service becomes a claimed diploma on a multilingual learner profile, which misleads future educators and employers.
Worked acceptance test. Create sample records for invited, enrolled, attended and completed states. Check which are allowed by the actual source property and flag unrepresented states. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
7. Assessment scale
A mathematics result is reported on a local 1–6 scale. A receiving dashboard labels it ‘percentage score’ because the interface expects 0–100. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Preserve source scale, code and meaning before localizing descriptive bands. A target-language label must tell readers whether they see a raw score, scaled result, proficiency level or percentage. A high performance band is rendered as 6%, manufacturing a dramatic decline that the assessment never measured.
Worked acceptance test. Test source values at the minimum, maximum and threshold boundaries. Confirm no numeric conversion occurs without a documented rule. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
8. Missing versus not applicable
An imported form contains blank results, an explicit ‘not assessed’ code and a ‘not applicable’ code. The target-language view renders all three as a dash. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Retain distinct missing-data states and expose their meaning in help text. Blank may reflect a technical failure, a collection gap or a permitted omission; it should not become a zero. Analysts count unassessed learners as failed learners, and teachers receive inappropriate intervention alerts.
Worked acceptance test. Construct one record for each missing state and one legitimate zero. Validate aggregate counts and accessible labels. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
9. Education organization identifier
Two schools share a translated short name but hold different official organization IDs. A public report localizes the ID column header as ‘school name’. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Keep identifiers, legal entity names and public display names in separate fields. Explain whether the code identifies a campus, an operator or an administrative organization. A funding report joins the two schools on a visible name and silently combines their totals.
Worked acceptance test. Use similarly named campuses and an institution that rebrands. Verify historical organization IDs and effective dates remain stable. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
10. Parent, guardian and contact
An early-learning system stores legally responsible guardians and separate emergency contacts. A translator labels every related person ‘parent’. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Translate the actual relationship category and describe its operational purpose. The authorized contact for an emergency is not automatically the legal decision-maker for educational consent. A multilingual form may send private material to someone with no right to receive it, or exclude a lawful guardian from communication.
Worked acceptance test. Check synthetic records with two guardians, one emergency contact and a changed custody arrangement. Review permissions separately from labels. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
11. English learner classification
A report contains a formal language-support category defined by local policy. A literal translation suggests a judgement of intelligence rather than a service eligibility rule. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Give the source policy definition in plain language, identify the relevant assessment or decision and use respectful non-stigmatizing wording. Do not infer cultural background from the code. A translated dashboard makes a support category appear to be an intrinsic characteristic of the learner.
Worked acceptance test. Ask an educator and language specialist to review the term with a neutral synthetic example. Verify the official category remains intact. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
12. Disability accommodation record
A course platform lists a documented accommodation and another system lists a diagnosis. A translated export merges both under ‘special needs’. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Distinguish the approved learning support arrangement from any health information. Translate what an institution is authorized to share for the specific operational task. Sensitive personal information may be exposed unnecessarily and accommodations may be misapplied.
Worked acceptance test. Run permission tests for teacher, administrator and external vendor roles. Confirm the localized report minimizes data as the original does. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
13. Learner mobility history
A student transfers twice in one year. A data warehouse stores enrollment intervals and distinct schools but a simple translation suggests a single current institution. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Describe start and end dates, enrollment relationship and source provenance. Readers must be able to distinguish the present arrangement from a historical one. An analyst assigns attendance at a former school to the receiving school, distorting both institutions’ rates.
Worked acceptance test. Check boundary dates, overlapping provisional registrations and retroactive corrections without rewriting original history. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
14. Early-learning age bands
An early-childhood programme describes age bands according to eligibility dates. A translation substitutes generic words such as toddler and preschooler. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Use the programme’s exact age interval or eligibility criterion as the definition. Familiar developmental terms can assist readers but must not replace the formal boundary. Parents are incorrectly told a child is ineligible for a service because the translated category uses a different age convention.
Worked acceptance test. Verify dates of birth on either side of the cutoff and a child whose admission decision used an exception. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
15. Qualification versus competency
A data map connects a course outcome to a competency descriptor. The target-language version calls the competency ‘a certificate’. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Separate a learned or assessed capability from a credential issued by an authorized body. Competency mapping can inform award criteria but does not itself grant the award. Learners believe they received a formal credential when the record only describes a skill target.
Worked acceptance test. Compare assessed, not assessed and credential-awarded sample cases. Check that the award issuer and date remain explicit. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
16. Employment outcome reporting
A workforce dataset tracks employment status after an educational programme. One localized heading says ‘career success’, which is broader than the recorded evidence. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Translate the exact employment measure, measurement date and survey population. Explain whether an outcome was self-reported, administratively matched or inferred. Public readers mistake a narrow employment indicator for proof of job quality, earnings growth or causal effect.
Worked acceptance test. Test records with part-time employment, unknown status and multiple jobs. Verify denominator and response-rate notes appear. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
17. Learner’s course load
A university describes study load using credits, contact hours and full-time-equivalent measures. A translation makes all three sound like time spent attending lectures. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Use distinct glossary entries for institutional credits, scheduled contact hours and calculated enrollment intensity. Preserve the source credit system. A cross-border adviser equates credits directly and incorrectly concludes that a learner has met a graduation requirement.
Worked acceptance test. Present two programmes with different credit definitions; require a reviewer to explain each measurement unit. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
18. School census snapshots
An administrator exports a census-date headcount and a live enrollment feed. A translated chart combines the two under ‘current students’. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Name the observation date and source system. A point-in-time count and live register answer different questions even when both concern students. The public report appears internally contradictory because users cannot see that the numbers refer to different moments.
Worked acceptance test. Compare changes one day before and after the census. Prove both charts label the time basis. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
19. Option code changes
A new CEDS release changes or adds a definition and an institution keeps an older local label. Both are displayed as though approved for Version 13. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Record the actual property and option-set version attached to each translation. Do not silently reinterpret historical records using a later definition. Reports become impossible to reproduce because old figures are retroactively narrated under different categories.
Worked acceptance test. Open archived and current extracts together. Confirm each description resolves to its applicable version. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
20. Multiple languages for one label
The same property has English, Spanish, Arabic and Mandarin labels. One translation reflects a narrower notion of school completion than the others. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Maintain one source definition and language-specific explanatory notes where a simple label is ambiguous. Translation memory should track semantic review, not just word agreement. Dashboards in different locales filter the same data differently because label-driven logic has been allowed into the reporting layer.
Worked acceptance test. Switch locale without changing data rows or filters. Ask bilingual reviewers to reconstruct the same operational rule. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
21. Numeric data and decimal conventions
An early-learning report displays a fraction of pupils participating in a programme. A localized interface changes commas and decimal points. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Format numbers for readers but retain numeric values and units in structured fields. Tell readers whether values are proportions, percentages or counts. A figure of 0.75 becomes seventy-five learners rather than a proportion, or a thousands separator is mistaken for a decimal point.
Worked acceptance test. Test values with zeros, large counts and decimals. Compare machine values before and after display formatting. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
22. Sex, gender and reporting categories
A statutory education return has a controlled reporting field whose categories differ from a school’s inclusive communication vocabulary. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Document the exact purpose and legal reporting definition while using respectful display wording. Do not infer how individuals identify from an administrative category. A generic translated word may conceal that a particular field is restricted, optional or mandated in one jurisdiction.
Worked acceptance test. Review legal and privacy notes with authorized specialists. Test missing and withheld categories explicitly. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
23. Race and ethnicity fields
One dataset uses locally legislated reporting categories that do not match another nation’s social categories. Translating category names as direct equivalents is tempting. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Explain the origin and limited applicability of the source categories. Do not force a person into another country’s classification system through translation. Cross-national comparisons create misleading or discriminatory conclusions and strip away the original survey context.
Worked acceptance test. Include a no-equivalence outcome in the mapping table and require methodological approval before aggregation. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
24. Address and geolocation
A family’s postal address includes non-Latin script and an administrative district name. A data exchange wants a romanized label for search. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Separate source address, display transliteration, postal code and geographic identifier. A transliteration may improve navigation but cannot overwrite the legal or deliverable address. A delivery system routes a letter to the wrong region because the localized display has replaced canonical address elements.
Worked acceptance test. Use a synthetic address with multiple accepted spellings and a stable locality code. Check that no fields move across components. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
25. Reporting definitions and aggregation
An agency calculates a four-year completion rate, while a school publishes the number of students finishing this year. A translation calls both ‘graduation rate’. The source field is the reference point. Before choosing a fluent translation, inspect the property definition and the role it plays in the complete data relationship. A useful target label is one that allows a reader to make the same decision the source data permits, not a label that sounds comfortable in isolation.
Explain numerator, denominator, cohort start and follow-up period, including exclusions. A metric definition is not portable merely because its title is short. Policy readers infer the two rates are identical and draw unsupported performance comparisons.
Worked acceptance test. Recompute sample reports from clearly labeled inputs. Flag every place a cohort definition or exclusion differs. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
26. Corrections and provenance
A registrar corrects a duplicated learner record. The local-language report should display the corrected state without hiding that a correction took place. A table can retain valid codes while describing them incorrectly. Therefore a successful test asks both whether the underlying data remained intact and whether a real reader would infer the correct condition from the translated explanation. These are separate checks, and both are necessary.
Preserve source event history, timestamps, authoritative correction status and the relationship between versions. Translate user-facing audit labels accurately. A reviewer cannot understand why yesterday’s total changed, or interprets a correction as evidence that a student withdrew.
Worked acceptance test. Test before-and-after views, a retroactive correction and access-controlled audit fields. Confirm the explanatory text matches the actual change. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
27. Open-source model reuse
An international project adopts a CEDS-aligned logical model but maintains several local extensions. A translated document simply declares ‘fully standardized’. A local policy may use a near-synonym with a different threshold. The professional response is to explain the mismatch, attach an explicit mapping status and let the responsible education authority decide whether transformation is permitted. The translator contributes semantic evidence, not policy authority.
Describe which elements are directly aligned, locally extended or still unmapped. Naming an alignment does not create implementation conformance by itself. A partner assumes unsupported data will exchange automatically and signs off an integration that loses fields.
Worked acceptance test. Publish a coverage matrix with explicit partial mappings, validation evidence and outstanding decisions. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
28. Publishing a public glossary
A translated education data glossary is intended for parents as well as analysts. Simplification is necessary, but a one-word summary removes qualifications from formal definitions. Whenever the source definition is narrower than an everyday word, place the boundary in the visible explanation. Short labels may be concise, but hover text, glossary entries and report notes should help the reader understand eligibility, time period and provenance without guessing.
Offer a short label, a plain-language explanation and a route to the authoritative definition. Help text should say what a field measures and what it cannot show. Readers begin to treat the simplified phrase as the legal meaning, and training teams copy it into unrelated policy documents.
Worked acceptance test. Ask a parent, teacher and data steward to answer the same example question after reading each language version. Correct ambiguity before release. The reviewer should record whether the display is faithful, whether machine identifiers and allowable values are unchanged, and whether the wording makes uncertainty visible. A failed test is an opportunity to fix the label, explanation or documented crosswalk, not to massage data until a translation appears to fit.
Four-week learning and implementation plan
Week 1 — Inventory and understand
Collect property IDs, definitions, versions, option sets, existing source labels, transformation rules and reporting documents. Select a manageable pilot domain. Identify which terms are legal or administrative categories, and which merely describe user-interface language. Produce a source-meaning register before translating a single heading.
Week 2 — Translate and annotate
Draft target-language labels alongside longer descriptions and worked examples. Have an education practitioner identify locally ambiguous categories and a technical reviewer protect identifiers, types and codes. Classify each mapping as direct, qualified, computed or unsupported rather than forcing artificial equivalence.
Week 3 — Validate with representative records
Use fabricated records to test boundaries: unknown versus zero, academic-year rollover, grade versus age, institution versus campus, current versus historical status. Compare calculations and group counts under two language views. Ensure accessibility labels, tables and help text retain the original constraints.
Week 4 — Publish with governance
Approve translations, record reviewers and effective versions, publish the glossary and an escalation path, and plan periodic review when the standard or local policy changes. Check that families can read the plain-language layer and that data stewards can still trace every published term back to the unchanged source definition.
Frequently asked questions
Can a translated CEDS label replace the original property identifier?
No. Keep the authoritative property identity intact and store the target-language label as presentation metadata or approved supporting documentation. This permits a system to display several languages while making the same queries.
Does CEDS make countries’ grades or qualifications equivalent?
No. It provides shared definitions and modeling resources. Policy equivalence, recognition of qualifications and eligibility decisions require additional authoritative rules, evidence and often human review.
What if an option-set label has no precise equivalent?
Keep the code, provide a plain-language explanation and record a non-equivalence or qualified mapping rather than hiding the distinction under a convenient but misleading term.
Which CEDS version should a project use?
Use the version declared by the source implementation and its governing data contract. Version 13 is published, but some operational systems may still rely on earlier versions. Document the choice.
Should translators edit numeric values or codes?
Not to make them seem locally familiar. Number formatting in a display is different from transforming a structured value. Any real transformation requires an explicit approved rule and reproducible test.
How do we avoid stereotyping in educational categories?
State what the field technically means, who defined it and which use it supports. Prefer respectful and exact terminology, avoid inferring identity from a record and include limitations where categories are policy-specific.
What proves that a localization is correct?
Evidence includes a versioned source baseline, bilingual semantic review, code and schema checks, representative boundary-case tests, documentation of exceptions and sign-off by the appropriate data owner.
How does this guide relate to OneRoster and CASE?
CEDS concerns definitions and data models; OneRoster exchanges school rosters and gradebook-related information; CASE exchanges competencies and standards. Their data may be mapped, but their technical responsibilities are different.
Sources and next reading
Use the official CEDS Version 13 property downloads, the CEDS data dictionary guidance and the CEDS data model to verify exact definitions. For a complementary school roster standard read OneRoster 1.2; for standards alignment read CASE 1.1. Return to the Translate Like a Pro series index.