WHY ENGLISH?
Make a dataset interpretable by people and machines with precise field definitions, data types, units, code lists, missing-value rules, provenance, stewardship and version history
Writing a data dictionary means explaining what each field means, how it is stored, which unit or format applies, which values are allowed, how codes map to labels, what missing values mean, where the data came from, who owns the definition and which version is current. English matters because “score”, “attendance”, “date” and “status” are not self-defining.
Current NIH materials say a human-readable data dictionary should include, at minimum, variable names, descriptions or labels, variable types, and response options and codes. NIDDK’s guidance similarly highlights names, labels, types and corresponding code lists. These are foundations: a real project may also need units, ranges, missing-data rules, provenance, validation, privacy classification and change history.
This guide follows a fictional school club participation dataset. It does not publish real student records, prescribe one metadata standard, guarantee interoperability, repair inaccurate source data or authorise access. Personal and sensitive data require appropriate governance, minimisation and approved systems beyond a dictionary.
Article contents
Open all chapters
- 1. Inventory the Actual Data Assets
- 2. Use Stable Field Names and Human Labels
- 3. Write Definitions With Boundaries
- 4. Specify Data Type and Storage Format
- 5. State Units, Precision and Reference Frame
- 6. List Allowed Values and Code Meanings
- 7. Define Missing, Unknown and Not Applicable Separately
- 8. Record Source and Derivation
- 9. Document Validation and Quality Notes
- 10. Assign a Steward and Review Route
- 11. Version Material Changes Transparently
- 12. Test With Records and a New Reader
ORIENTATION
What this guide covers
Current fact check
Checked on 9 November 2026, NIH/NIEHS guidance stated that a data dictionary should include at least variable names, descriptions or labels, variable types, and response options with their codes. Current NIDDK data-dictionary guidance likewise described human-readable dictionaries through variable names, labels or descriptions, types and code lists. NCBI’s dbGaP submission guide illustrates that coded values such as 1, 2, 3 and 99 need defined meanings. These examples support clear metadata but do not establish one universal schema.
What this guide does not cover
This article owns planning, writing, validating, reviewing and versioning a human-readable data dictionary. It does not own data collection, consent, database design, statistical analysis, privacy impact assessment, access control, retention, de-identification, data cleaning, a universal ontology, an API contract or permission to disclose records.
Worked example
Every chapter returns to a fictional school club participation dataset with member IDs, activity dates, session type, duration, attendance status, consent flag and a reason code for an excused absence. It is fictional, low-stakes and educational; it does not report a real school, clinical decision or guaranteed outcome.
Did You Know? For data dictionary, precise communication makes appraisal possible but does not perform the appraisal, remove bias or guarantee transfer.
Warm up with eduKate’s guides to explaining scientific evidence and asking better questions in class.
CHAPTER 1 OF 12
1. Inventory the Actual Data Assets
Why this move matters
List files, tables, sheets and extracts, then choose which fields and version the dictionary describes.
Did You Know? Scope ties definitions to real data.
Treat every field as a promise of meaning
List files, tables, sheets and extracts, then choose which fields and version the dictionary describes. In The project distinguishes the attendance table from the member profile table and names the 2026-11 export., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: A dictionary for an imagined future schema can mislead current users.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. A dictionary for an imagined future schema can mislead current users. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: List files, tables, sheets and extracts, then choose which fields and version the dictionary describes.
- Boundary: A dictionary for an imagined future schema can mislead current users.
- Application: The project distinguishes the attendance table from the member profile table and names the 2026-11 export. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Inventory the Actual Data Assets” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 2 OF 12
2. Use Stable Field Names and Human Labels
Why this move matters
Record the stored name exactly and pair it with a readable label without casually renaming columns.
Did You Know? Stable identity supports joins and audit.
Treat every field as a promise of meaning
Record the stored name exactly and pair it with a readable label without casually renaming columns. In attn_status is labelled “Attendance status” while retaining the exact database field name., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: A friendly label and a machine field name serve different purposes.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. A friendly label and a machine field name serve different purposes. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Record the stored name exactly and pair it with a readable label without casually renaming columns.
- Boundary: A friendly label and a machine field name serve different purposes.
- Application: attn_status is labelled “Attendance status” while retaining the exact database field name. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Use Stable Field Names and Human Labels” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 3 OF 12
3. Write Definitions With Boundaries
Why this move matters
State the concept, population, event and exclusions needed to interpret the field.
Did You Know? A definition says what belongs and what does not.
Treat every field as a promise of meaning
State the concept, population, event and exclusions needed to interpret the field. In Session duration means scheduled minutes for one club session, excluding travel and optional social time., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: Circular definitions such as “status of attendance” add no meaning.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. Circular definitions such as “status of attendance” add no meaning. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: State the concept, population, event and exclusions needed to interpret the field.
- Boundary: Circular definitions such as “status of attendance” add no meaning.
- Application: Session duration means scheduled minutes for one club session, excluding travel and optional social time. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Write Definitions With Boundaries” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 4 OF 12
4. Specify Data Type and Storage Format
Why this move matters
Name text, integer, decimal, Boolean, date or other type plus the stored format where relevant.
Did You Know? Type and format prevent silent conversion.
Treat every field as a promise of meaning
Name text, integer, decimal, Boolean, date or other type plus the stored format where relevant. In activity_date is an ISO-style date in YYYY-MM-DD, not free text., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: Display formatting can differ from stored representation.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. Display formatting can differ from stored representation. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Name text, integer, decimal, Boolean, date or other type plus the stored format where relevant.
- Boundary: Display formatting can differ from stored representation.
- Application: activity_date is an ISO-style date in YYYY-MM-DD, not free text. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Specify Data Type and Storage Format” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 5 OF 12
5. State Units, Precision and Reference Frame
Why this move matters
Name minutes, metres, percentages, currency or another unit and explain rounding or reference period.
Did You Know? Units turn magnitude into meaning.
Treat every field as a promise of meaning
Name minutes, metres, percentages, currency or another unit and explain rounding or reference period. In duration_min is an integer number of scheduled minutes for the session., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: A number without a unit can be unusable.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. A number without a unit can be unusable. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Name minutes, metres, percentages, currency or another unit and explain rounding or reference period.
- Boundary: A number without a unit can be unusable.
- Application: duration_min is an integer number of scheduled minutes for the session. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “State Units, Precision and Reference Frame” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 6 OF 12
6. List Allowed Values and Code Meanings
Why this move matters
Provide the full code list, labels, case rules and any valid range.
Did You Know? Code lists make categories readable.
Treat every field as a promise of meaning
Provide the full code list, labels, case rules and any valid range. In P means present, A absent, E excused and C cancelled; lowercase values are invalid., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: The code 0 does not automatically mean no or missing.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. The code 0 does not automatically mean no or missing. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Provide the full code list, labels, case rules and any valid range.
- Boundary: The code 0 does not automatically mean no or missing.
- Application: P means present, A absent, E excused and C cancelled; lowercase values are invalid. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “List Allowed Values and Code Meanings” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 7 OF 12
7. Define Missing, Unknown and Not Applicable Separately
Why this move matters
Explain blanks, nulls and reserved codes so absence of a value is not mistaken for zero.
Did You Know? Missingness is data about the collection process.
Treat every field as a promise of meaning
Explain blanks, nulls and reserved codes so absence of a value is not mistaken for zero. In Null means not recorded, while N/A means the field does not apply to a cancelled session., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: Using 99 as missing must be documented and kept outside valid ranges.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. Using 99 as missing must be documented and kept outside valid ranges. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Explain blanks, nulls and reserved codes so absence of a value is not mistaken for zero.
- Boundary: Using 99 as missing must be documented and kept outside valid ranges.
- Application: Null means not recorded, while N/A means the field does not apply to a cancelled session. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Define Missing, Unknown and Not Applicable Separately” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 8 OF 12
8. Record Source and Derivation
Why this move matters
Name the originating system or form and explain any calculation or transformation.
Did You Know? Provenance helps users judge fitness for purpose.
Treat every field as a promise of meaning
Name the originating system or form and explain any calculation or transformation. In attendance status comes from the check-in form; participation rate is derived from eligible attended sessions divided by eligible scheduled sessions., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: A derived field should not masquerade as directly observed.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. A derived field should not masquerade as directly observed. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Name the originating system or form and explain any calculation or transformation.
- Boundary: A derived field should not masquerade as directly observed.
- Application: attendance status comes from the check-in form; participation rate is derived from eligible attended sessions divided by eligible scheduled sessions. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Record Source and Derivation” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 9 OF 12
9. Document Validation and Quality Notes
Why this move matters
State uniqueness, range, relationship or consistency checks and known limitations.
Did You Know? Quality notes prevent false confidence.
Treat every field as a promise of meaning
State uniqueness, range, relationship or consistency checks and known limitations. In member_id plus activity_date plus session_id must be unique, but two legacy months have incomplete reason codes., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: A validation rule does not prove the source event was true.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. A validation rule does not prove the source event was true. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: State uniqueness, range, relationship or consistency checks and known limitations.
- Boundary: A validation rule does not prove the source event was true.
- Application: member_id plus activity_date plus session_id must be unique, but two legacy months have incomplete reason codes. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Document Validation and Quality Notes” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 10 OF 12
10. Assign a Steward and Review Route
Why this move matters
Name the role that owns the definition, approves changes and answers interpretation questions.
Did You Know? Ownership keeps metadata alive.
Treat every field as a promise of meaning
Name the role that owns the definition, approves changes and answers interpretation questions. In The activities data steward reviews status-code changes; coaches submit requests through the maintained form., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: An owner should be a durable role, not only one person’s inbox.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. An owner should be a durable role, not only one person’s inbox. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Name the role that owns the definition, approves changes and answers interpretation questions.
- Boundary: An owner should be a durable role, not only one person’s inbox.
- Application: The activities data steward reviews status-code changes; coaches submit requests through the maintained form. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Assign a Steward and Review Route” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 11 OF 12
11. Version Material Changes Transparently
Why this move matters
Record effective dates and changes to definitions, types, units, codes or derivations.
Did You Know? Versioning protects time-based meaning.
Treat every field as a promise of meaning
Record effective dates and changes to definitions, types, units, codes or derivations. In Version 1.2 adds code C for cancelled and explains that earlier blanks may include cancellations., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: Overwriting history can make old analyses impossible to reproduce.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. Overwriting history can make old analyses impossible to reproduce. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Record effective dates and changes to definitions, types, units, codes or derivations.
- Boundary: Overwriting history can make old analyses impossible to reproduce.
- Application: Version 1.2 adds code C for cancelled and explains that earlier blanks may include cancellations. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Version Material Changes Transparently” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
CHAPTER 12 OF 12
12. Test With Records and a New Reader
Why this move matters
Ask someone outside the build team to interpret sample rows and run the stated rules.
Did You Know? Successful interpretation is the practical test of a dictionary.
Treat every field as a promise of meaning
Ask someone outside the build team to interpret sample rows and run the stated rules. In A new analyst correctly distinguishes zero minutes from an unrecorded duration and identifies one invalid code., connect the stored field name to a human definition, type, format, unit, code list and missing-data rule.
Distinguish labels, codes and actual values
Show how values are represented and what each permitted code means. Keep this boundary: Expert familiarity can hide ambiguous wording.
Record provenance, ownership and change
Name the source, steward, effective version and material revisions so users can interpret the dataset that actually exists.
Test the dictionary against real records and users
Ask a new analyst to interpret sample rows, run validation rules and flag ambiguous terms. Expert familiarity can hide ambiguous wording. A dictionary supports reuse; it does not repair poor collection or authorise access to restricted data.
Practical reader check
- Evidence: Ask someone outside the build team to interpret sample rows and run the stated rules.
- Boundary: Expert familiarity can hide ambiguous wording.
- Application: A new analyst correctly distinguishes zero minutes from an unrecorded duration and identifies one invalid code. Name the next output and how it will be checked.
Student, parent and career transfer
Practise “Test With Records and a New Reader” by separating evidence, interpretation, decision and next action. A parent, teacher or teammate can ask what supports the sentence and which missing fact could change it. This discipline transfers to research, education analytics, public administration, health data, finance, product analytics, databases, surveys, spreadsheets, reporting and machine-learning projects without guaranteeing grades, accessibility conformance, software fixes, interoperability, adoption, approval, employment or future success.
PRACTICE LAB
Build a data dictionary decision trail
Student route
- Choose one real example or official document inside this guide’s scope.
- Copy its objective and underline actor, action, context, evidence and intended decision.
- Draw the specific reasoning chain used in this article.
- Mark each step reported, inferred, contested or unknown.
- Rewrite the conclusion so it preserves the narrowest important limitation.
Parent or teacher route
Ask the learner to explain make a dataset interpretable by people and machines with precise field definitions, data types, units, code lists, missing-value rules, provenance, stewardship and version history twice: once to a classmate and once to a method specialist. Praise a precise correction and a well-stated unknown; both show disciplined attention to evidence.
Early-career route
Create a one-page data dictionary audit with source link, date checked, scope, method logic, decisive wording, contrary evidence, limit and next action. A colleague should be able to continue without guessing.
Decision rule
If a shorter explanation removes the condition that made the original claim true, it is not clearer yet. Restore the condition.
Continue with eduKate’s guides to reading course and career information and giving useful feedback in group projects.
USEFUL NEXT READING
Connect this narrow owner to its neighbour
Read the education-indicator metadata guide for a related but different reader job. Return here when the question is specifically how to make a dataset interpretable by people and machines with precise field definitions, data types, units, code lists, missing-value rules, provenance, stewardship and version history. Use each guide for the reader task named in its heading.
For English practice across research, school and career communication, visit the Singapore English Tuition Centre learning site.
FREQUENTLY ASKED QUESTIONS
Questions students and parents often ask
Is a data dictionary the same as a glossary?
No. A glossary defines concepts broadly; a data dictionary ties definitions to actual fields, types, values and rules.
What is the minimum useful content?
Current NIH guidance highlights variable names, descriptions or labels, types, and response options with their codes.
Should blank, zero and “not applicable” be the same?
Usually not. Define each state explicitly because they can support different interpretations.
Can a spreadsheet serve as a data dictionary?
Yes if it is controlled, readable, versioned and connected to the actual dataset; the file format alone does not ensure quality.
Who should approve changes?
Assign a steward or governance role familiar with both the subject meaning and technical consequences.
Does a dictionary make data safe to share?
No. Access, consent, privacy, licensing, security and retention still require separate controls.
PRIMARY SOURCES
Read the primary guidance for this data dictionary
Facts were checked against current official and primary sources on 9 November 2026. Follow the live source for later corrections, completed updates or companion material.
- NIEHS data-dictionary guidance
- NIDDK preparing data dictionaries
- NCBI dbGaP submission guide
- eduKate education-indicator metadata guide
- eduKate data, graphs and charts guide
A cheerful final note
Precise English makes make a dataset interpretable by people and machines with precise field definitions, data types, units, code lists, missing-value rules, provenance, stewardship and version history more generous, not less. People can see how their evidence and values entered the work, where disagreement remains and what a responsible next step looks like.