How do you translate zero, null, blank, unknown, missing, not applicable, not recorded and withheld without changing what a document actually says? This complete guide to missing-value translation and data-label localization explains how to preserve the distinction between a genuine zero and the absence of information, between an unanswered field and an inapplicable field, and between an unknown value and a deliberately concealed one. Translating a missing-data label accurately is not merely finding a natural-sounding synonym. It is preserving what the sender knows, what the sender does not know, what happened in the data-collection process, and what the reader may reasonably conclude.
The hidden problem in translating forms, surveys, dashboards, spreadsheets, laboratory reports, interfaces, databases and technical documents is that an empty-looking field may carry several incompatible meanings. A value can be zero, absent, null, not asked, unanswered, declined, pending, unavailable, unreported, deleted, unsupported, redacted, estimated, suppressed or not applicable. These distinctions affect calculations, eligibility, privacy, audit trails, clinical records, research findings, customer support and public trust. This chapter gives translators, localisation teams, content editors, data stewards and reviewers a practical system for carrying those distinctions across language boundaries without inventing facts.
If you are searching for how to translate missing data, whether null means zero, how to localize N/A or unknown, how to translate a blank field, how to keep a masked value distinct from an absent one, or how to review AI-translated forms, begin with the source-data state rather than the English display label. This world-facing guide is EDKSG-TRANS-MASTER-WORLD-1810 in Master Art of Translation | The Complete System for Moving Meaning Between Languages. It follows the Measurement and Uncertainty Fidelity System, which owns numerical measurements and rounding; this chapter owns the meaning of the missing value itself, not the numerical quantity.
Quick answer: First identify whether a value exists; then identify whether its absence is semantic, procedural, technical, privacy-related, temporary, or deliberate. Preserve the exact distinction in the target language and its data representation. Keep human display text separate from controlled machine codes. Test zero, blank, null and every applicable absence reason as different cases. Never turn an unmeasured quantity into a measured zero, a withheld fact into a nonexistent fact, or a pending result into a negative result. The examples below are invented editorial exercises, not real patient, financial or customer records.
1. The decision the reader needs before the terminology
A person opening a translated form often does not care whether the original English word was unavailable or missing in isolation. The reader wants to know whether to supply information, wait for a result, skip a field, appeal a decision, ask for clarification or assume that a quantity is zero. Those actions are different. A good translation keeps the choice space intact. When the source is ambiguous, the translator should not fabricate certainty; the appropriate output can be a transparent editorial query, accompanied by a provisional label that does not overclaim.
Consider four rows in a service report. Row A says incidents: 0; row B says incidents: not recorded; row C says incidents: not applicable; row D says incidents: pending verification. Only the first row communicates an actual reported zero. The second lacks a record, the third does not belong to the measurement’s conceptual domain, and the fourth refers to an incomplete workflow. Replacing every label with a blank dash may create a clean-looking table, but the reader can no longer understand what happened. The translation must preserve these states even when design space is limited.
A second reader may not see the table at all. It may be consumed by a screen reader, an exported spreadsheet, an API, or a dashboard that turns text labels into colours and totals. Good localisation therefore tests the visible wording, the structured value and the behaviour of each downstream interface. Translators need not become database engineers, but they must know when a label is a presentation choice and when it is an identifier, code, computation input or security boundary.
2. What counts as missing-value meaning in translation?
1. Measured zero
A source value of 0 is a claim about quantity: an observed count, computed result or stored value. It is not a placeholder for anything the source lacks. Preserve whether it is exact, rounded, reported or estimated, and do not infer that zero events were observed if the collection process did not operate. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
2. None reported
No cases were reported is about reporting events, not necessarily about actual absence. A translator must distinguish a reporting-channel count from an observation about the world. If reports could be incomplete, do not rewrite the sentence as no cases occurred. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
3. Empty visible cell
A blank cell is a presentation condition. Without metadata, it does not identify why the data is absent. The translator should retrieve a dictionary, form instruction or export mapping before choosing a semantic term. If none exists, do not create a false explanation. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
4. Explicit null
In JSON and similar systems, null is a specific representational value, not a synonym for zero, an empty string, a false boolean or an omitted property. The interface may display it as a localized phrase, but the underlying value must remain intact when the specification requires it. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
5. Omitted property
A field can be entirely absent from a message. That absence may have a different contractual or technical effect from an included field with null. Schema rules determine whether it is permitted. A translation cannot safely replace the omitted property with null simply for visual symmetry. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
6. Unknown
An expected value may exist but be unknown to the current source. This is a statement about knowledge, not necessarily about measurement, reporting or privacy. Ask whether the value is expected to be discoverable and whether later updates are planned. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
7. Asked but unknown
The respondent was asked and could not provide a value. The event of asking matters. A form translated as not asked would misrepresent collection effort and might trigger unnecessary follow-up. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
8. Not asked
A workflow did not request the value. That cannot be converted into a refusal or a data-loss claim. The translation should preserve whether the omission was designed, conditional or accidental when the source supports such precision. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
9. Declined to answer
A person was asked and chose not to disclose. This is a refusal or choice not to answer, not a cognitive inability to know the answer. Respectful and accurate phrasing avoids implying deception, incapacity or noncompliance without evidence. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
10. Withheld
The holder intentionally does not release information. The reason may be privacy, security, contractual restriction or pending authorization; do not guess which. Translate the deliberate nonrelease without falsely claiming the value never existed. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
11. Masked
The stored value is concealed in the display, often by a controlled process. Masked differs from erased and from unavailable. A row showing masked identification characters should not be translated as an identifier with no recorded value. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
12. Redacted
A portion was removed or obscured in a released copy. A translator must preserve that the public version lacks content because of an editorial or disclosure action, rather than claiming the original document was empty. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
13. Not applicable
The field has no appropriate value for the case. It is not merely unanswered. The predicate itself fails to apply. Translation needs the domain’s definition: the same field might apply to another class of record. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
14. Not performed
A test, survey item, inspection or calculation was never carried out. It cannot legitimately be labelled normal, negative or zero. Distinguish nonperformance from a performed procedure that returned no result. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
15. Pending
The process has started but the reported value is not yet finalized. Pending expresses workflow state and can change with time. It is not a stable negative result or permanent missing record. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
16. Temporarily unavailable
A value is currently inaccessible, with an expectation or possibility that access may return. Translators must not promise recovery if the source only says temporarily unavailable. The timing and responsible party may remain unspecified. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
17. Not yet collected
A collection action is scheduled or has not taken place. Not yet encodes a temporal expectation relative to the current reporting moment. It must not silently become never collected. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
18. Lost
Information existed or was expected but became unavailable through loss. This is stronger than merely missing from a report. Preserve provenance and avoid attributing blame unless the source provides it. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
19. Deleted
A removal operation occurred. Deletion is a lifecycle event with access, versioning and recovery implications. It is not identical to a blank display, especially where audit logs preserve the operation. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
20. Unsupported
The source system cannot represent or provide this field. This is a capability limitation rather than a statement about the real-world value. Never translate unsupported as none or not applicable unless the specification says so. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
21. Error
A software, instrument or workflow failure obstructed the value. It signals a defect or interruption, not an actual measurement. Where codes are controlled, preserve the machine token while translating the accompanying human explanation. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
22. Not reported
The expected item was not included in a publication or communication. It might still exist in the underlying records. Translate reporting status without assuming data was never collected. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
23. Unavailable for disclosure
The record may exist but cannot be shared under current access rules. Translation must not disclose or infer its substance. A public explanation can be accurate without revealing private information. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
24. Imputed
A missing observation may be replaced by an analytical estimate produced under a documented method. The resulting number is not an observed fact. Preserve the marker and methodological caveat so the reader can distinguish estimation from measurement. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
25. Default value
A system may fill an omitted field using a default, such as a chosen preference or configured setting. That value was not necessarily explicitly selected by a person. Translation must not call it user-entered when it was system-generated. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
26. Below detection limit
A procedure may have been performed but unable to detect the quantity at the relevant threshold. Below detection is not the same as exact zero or unknown. Domain terminology and method details determine how the result should be stated. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
27. Suppressed small count
A table may intentionally hide counts to limit identification risk or protect confidentiality. The count can be nonzero. Translating the cell as zero could both distort statistics and mask the reason for suppression. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
28. Not evaluated
A criterion or case was not examined. A verdict such as pass, fail, compliant or noncompliant cannot be inferred from non-evaluation. Translation should preserve the evaluation state. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
29. No change
A recorded comparison may show no difference between two readings. That is not a missing reading. Preserve whether both observations exist and whether the contrast is meaningful or only a display convention. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
30. NaN or nonfinite computation
Technical numeric representations such as NaN or infinity are not ordinary prose labels. They may arise through calculation, not because a person skipped a field. Follow the data contract rather than turning them into colloquial unknown. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
31. No answer versus no response
A human may answer that they do not know, or may fail to respond at all. Surveys distinguish these events because they affect response rates, missingness analysis and the fairness of follow-up. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
32. No record found
A search returned no matching record in a defined system. This is bounded by dataset, identifiers and search conditions. It does not logically establish that an event or person never existed. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
33. Unverified
A value has been provided but has not passed verification. Do not erase the source claim or elevate it to a confirmed fact. Preserve the distinction between provisional evidence and absent evidence. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
34. Not yet released
A completed value or report can exist while dissemination is scheduled for later. A translation that says not available may hide this distinction; a translation that says not created is false. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
35. Out of scope
A dataset or review deliberately excludes a case. That is a boundary of design, not a missing value within the included population. Readers need to know which denominator or collection definition applies. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
36. Unspecified
A document leaves a detail undetermined. The author may not intend a particular answer, and the translator should not choose one. An unspecified time, role or parameter differs from unknown because no concrete value may be defined yet. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
37. Unconfirmed
The proposed value awaits confirmation. It can be true or false, but the claim has a different evidential status from a verified result. Markers such as tentative or unconfirmed cannot be dropped for smoother copy. Translation test: State what the reader may infer from this label, what the reader may not infer, and which source field or workflow record could independently settle its meaning. When a target-language shortcut merges these states, prefer a short explanatory phrase or an approved domain label over the shortcut.
3. The mechanism: six layers of a single apparently empty field
1. Existence layer
Does the underlying property have a value in the world, or is the question structurally inapplicable? An organization may know a value exists without knowing what it is; that is different from a property that makes no sense for this item. The translator should not resolve an ontological question from the shape of a blank field. In practice, record an evidence source for this layer: schema definition, form logic, interview protocol, technical log, legal disclosure rule, or the document’s own explanatory notes. A translator’s preference is not evidence.
2. Knowledge layer
Who knows the value: the subject, collector, system, reviewer, owner or reader? Statements of unknown status may be relative to a particular actor. A value absent from one dashboard can be fully known to another authorized team. An accurate target cannot convert local ignorance into universal nonexistence. In practice, record an evidence source for this layer: schema definition, form logic, interview protocol, technical log, legal disclosure rule, or the document’s own explanatory notes. A translator’s preference is not evidence.
3. Collection layer
Was the question asked, the procedure performed and the response captured? Some workflows branch around fields, some encounter refusal, and some fail technically. Those histories are different. The target label should keep the status needed for future follow-up, audit or analytic treatment. In practice, record an evidence source for this layer: schema definition, form logic, interview protocol, technical log, legal disclosure rule, or the document’s own explanatory notes. A translator’s preference is not evidence.
4. Representation layer
Is the value represented by a null token, omitted field, empty string, code, dash, blank, note or encrypted text? These are implementation choices with rules. Translate the human-readable label where appropriate; do not mutate schema keys, raw codes or numeric types while editing prose. In practice, record an evidence source for this layer: schema definition, form logic, interview protocol, technical log, legal disclosure rule, or the document’s own explanatory notes. A translator’s preference is not evidence.
5. Disclosure layer
Does a value exist but remain concealed because of privacy, security, embargo or contractual restrictions? Distinguish user-facing disclosure notices from the value itself. Correct translation must not weaken a protection rule or imply unauthorized access is possible. In practice, record an evidence source for this layer: schema definition, form logic, interview protocol, technical log, legal disclosure rule, or the document’s own explanatory notes. A translator’s preference is not evidence.
6. Decision layer
What should the next actor do? Zero may allow a calculation; pending may prompt waiting; refused may stop collection; not applicable may skip a validation rule; system error may require support. The semantic audit passes only if target readers would take a substantially equivalent next step. In practice, record an evidence source for this layer: schema definition, form logic, interview protocol, technical log, legal disclosure rule, or the document’s own explanatory notes. A translator’s preference is not evidence.
4. The repeatable source-to-target method: a 24-gate ledger
Gate 01 — Inventory the field
Record the exact source label, surrounding sentence, column name, unit, format, locale, version and target reader. Keep an unedited snapshot so later changes are traceable. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 02 — Identify the source owner
Find the author, system steward or governing vocabulary. Do not assume a public web definition outranks a project-specific controlled code list. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 03 — Distinguish fact from display
Separate the machine value and display phrase. A dash may be visual shorthand for a non-dash status; ask for the mapping. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 04 — Classify zero
Determine whether zero is an actual count, a rounded quantity, an instrument result or a placeholder. Never infer one from the other. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 05 — Classify inapplicability
Check whether the property can exist for the record at all. Record the conditional rule making the field inapplicable. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 06 — Classify collection status
Check whether information was solicited, observed, performed, recorded or published. Preserve each meaningful action. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 07 — Classify the knowledge state
Identify whether the owner lacks knowledge, the respondent lacks knowledge, or the value remains undetermined. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 08 — Classify refusal
Distinguish a choice not to answer from inability to answer, inaccessibility and system nonresponse. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 09 — Classify access restriction
Confirm whether a value is masked, redacted, withheld, embargoed or access-controlled. Do not infer a reason not supplied. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 10 — Classify time
Record whether the absence is pending, temporary, permanent, scheduled or unknown in duration. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 11 — Classify technical failure
Separate unsupported fields, parsing errors, transmission failures and legitimate non-values. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 12 — Check structured representation
Verify whether the source is null, absent, blank string, code, zero or nonfinite numeric marker. Treat the data contract as authoritative for the representation. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 13 — Confirm terminology
Use approved target terminology when a data dictionary or discipline-specific code set controls it. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 14 — Draft the reader-facing label
Choose the shortest natural phrase that preserves the state. Add an explanatory tooltip only if the visible label cannot carry the distinction. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 15 — Check a full sentence
A label may be ambiguous alone but precise in a sentence. Verify grammatical agreement, time, negation and scope in context. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 16 — Test screen and export
Inspect how the translation appears on mobile, in a PDF, in a spreadsheet, in a copied string and to assistive technology. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 17 — Test computational effect
Confirm zero is not excluded as missing, missing is not counted as zero and suppression is not converted into false data. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 18 — Test the negative inference
Ask what the label forbids the reader from concluding. Document the strongest tempting but unsupported inference. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 19 — Test the operational decision
Give the translated phrase to a reviewer who has not seen the source. Can they choose the correct next action? Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 20 — Check cultural implications
Avoid stigmatizing language for refusals and unknown answers. Keep the original status without adding moral judgment. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 21 — Review privacy and disclosure
Confirm that translations do not expose protected content, broaden consent or suggest that hidden values are freely available. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 22 — Review change history
If the status is corrected or updated later, verify that translated copies preserve chronology and distinguish superseded states. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 23 — Record uncertainty
When the source does not distinguish states, escalate an editorial query. A cautious label is not permission to invent metadata. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
Gate 24 — Authorize release
Complete independent review, fix accuracy issues, record any accepted limitations and verify the final rendered product, not only the drafting file. Pass evidence: a reviewer can identify the relevant input, the exact claim retained, one plausible unsafe alternative, and the observation that would falsify the chosen translation. This makes quality control reproducible instead of dependent on whether the phrasing merely sounds idiomatic.
5. How to use authoritative terminology without copying a codebook blindly
The JSON Schema documentation on null distinguishes a literal null from an absent property; RFC 8259 formally defines JSON null as a value alongside strings, numbers and booleans. These are software representation rules, not translations of every human-facing missing-data label. A UI can render an appropriate local-language explanation while preserving the underlying schema-defined value.
The HL7 FHIR DataAbsentReason value set illustrates why one generic missing label is often inadequate: codes distinguish unknown, asked but unknown, not asked, asked but declined, masked, not applicable, unsupported and process errors. This is cited here as a terminology example; health records must follow the applicable system version, implementation guide and professional governance rather than transplanting a label into another field.
The W3C Model for Tabular Data and Metadata on the Web describes an annotated semantic cell value and identifies a default mapping of an empty string to null within that model. The rule should not be generalized to every spreadsheet format or every operational table. A particular import pipeline may distinguish empty text, absent field and special sentinel codes differently. Translation quality depends on confirming the actual system’s definitions, not guessing from the visible CSV punctuation.
The Unicode CLDR numbers specification separates language-specific plural forms from explicit numeric-zero messaging and locale number formatting. A legitimate zero is a quantity, so pluralization and natural phrasing matter. A missing value is a different semantic state and should not be sent through the zero-message branch merely to simplify localization code.
For error review, the MQM Core typology distinguishes translation accuracy issues including omissions, additions and mistranslations from terminology, locale, linguistic and markup issues. A blank-to-zero conversion usually creates an accuracy problem even if a separate typography problem made it appear. Recording the error’s meaning and severity encourages reviewers to fix the decision risk rather than merely its surface.
Boundary with related eduKateSG chapters: the preceding Measurement and Uncertainty chapter governs numeric precision, units, tolerances and rounding. The translation root governs source analysis and the wider cross-language method. This chapter asks a different question: what does it mean when a supposedly available value is absent, unknown or intentionally concealed? It therefore does not replace vocabulary instruction, grammar instruction, statistical methodology or professional domain rules.
6. Failure modes: how a tiny missing-value word changes the reader’s world
The following error taxonomy is deliberately more specific than a generic list of bad synonyms. Each case identifies a source fact, an unsafe translation, its practical consequence and the correction to make during a bilingual review. The language examples are semantic English glosses; actual target-language phrasing must be chosen by competent native-language and domain specialists. The goal is to protect the original evidence and the reader’s options, not to make every translation literally resemble English.
Failure 01 — Converting an unrecorded count to zero
Source circumstance: A shift report has no entry because the device was offline. Unsafe target: The localized chart displays 0, an apparently reassuring measurement.
Repair and diagnostic: Restore the absence or instrument-outage state. Check the device log and whether totals exclude this row; a zero implies monitoring was functioning. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 02 — Converting an observed zero to a blank
Source circumstance: An inspected batch has zero documented defects. Unsafe target: The target table leaves the cell empty to avoid an awkward numeral.
Repair and diagnostic: Keep the counted zero and its inspection basis. Blank may be treated as incomplete or omitted, which can hide evidence of a completed inspection. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 03 — Replacing not applicable with unknown
Source circumstance: A field for an optional attachment is inapplicable because the record type cannot include one. Unsafe target: The translated form suggests staff have not yet found the attachment.
Repair and diagnostic: State that the field does not apply. Otherwise the system can create false work queues and pressure users to supply impossible information. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 04 — Replacing unknown with not applicable
Source circumstance: A purchaser’s eligibility date exists but the representative does not know it. Unsafe target: The target label indicates that eligibility dates are irrelevant.
Repair and diagnostic: Preserve the gap in knowledge. Calling the field inapplicable could skip a rule that still requires a date once known. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 05 — Treating not asked as refused
Source circumstance: A survey branch deliberately bypassed an age question for a respondent. Unsafe target: The translation marks the person as unwilling to answer.
Repair and diagnostic: Use the branch-status wording. A refusal implies an interaction that never happened and can alter nonresponse classifications. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 06 — Treating refusal as an unanswered field
Source circumstance: A respondent actively chooses the decline option. Unsafe target: The export translation says no response received.
Repair and diagnostic: Record declined, not mere silence. The distinction can affect consent handling, follow-up behaviour and survey interpretation. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 07 — Treating a masked value as absent
Source circumstance: An account identifier exists but is displayed as ending in four digits only. Unsafe target: The translated interface says no identifier is on file.
Repair and diagnostic: Use masked or partly hidden as specified. The record exists and its concealment should not be confused with missing registration. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 08 — Treating a redaction as deletion
Source circumstance: A public copy contains a blacked-out paragraph while the original is preserved under access control. Unsafe target: The caption says the paragraph was deleted from all records.
Repair and diagnostic: Describe the redacted public copy. Deletion would make a stronger claim about the source archive and might mislead a reviewer. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 09 — Turning pending into failed
Source circumstance: A qualification review is not yet complete. Unsafe target: A dashboard translates pending as unsuccessful.
Repair and diagnostic: Translate the undecided workflow state. A failed status implies an assessed negative decision that has not occurred. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 10 — Turning not performed into a negative test
Source circumstance: A test was omitted because a sample was damaged before processing. Unsafe target: A patient-facing explanation reads negative result.
Repair and diagnostic: Keep the procedure-not-performed status; no test outcome can be inferred. A qualified specialist must approve any clinical display wording. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 11 — Turning no match into no existence
Source circumstance: A document search finds no record for the supplied identifier. Unsafe target: A translated letter claims no such transaction occurred.
Repair and diagnostic: Preserve the scope of the search and its conditions. A search failure is evidence about that query, not proof about every possible record. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 12 — Turning suppression into zero
Source circumstance: A public table hides a small count to reduce identification risk. Unsafe target: The localized CSV sets the value to zero for easier import.
Repair and diagnostic: Preserve the suppression marker and disclosure rule. A fabricated zero changes totals and can hide the reason the cell was protected. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 13 — Turning below-detection into none
Source circumstance: A measurement is reported below the method’s detection limit. Unsafe target: The translator writes no substance present.
Repair and diagnostic: Retain the threshold relationship and method qualification. A value below the relevant threshold is not logically zero. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 14 — Turning out-of-scope into missing
Source circumstance: A report intentionally excludes an experimental group from a particular analysis. Unsafe target: The translated figure labels those records missing.
Repair and diagnostic: Describe the analysis boundary. Otherwise readers may blame collection failure and misunderstand which cases entered the denominator. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 15 — Turning not assessed into failure
Source circumstance: A vendor review left one criterion unevaluated. Unsafe target: The translated scorecard marks the criterion as unmet.
Repair and diagnostic: Preserve no assessment. An unmet criterion means evidence of noncompliance; unevaluated means the decision is unavailable. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 16 — Attributing a default to the user
Source circumstance: A settings screen fills a locale-dependent preference automatically. Unsafe target: The translated history says the user selected this value.
Repair and diagnostic: Keep the provenance as system default. Attribution matters when explaining why a setting appeared or changed without explicit choice. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 17 — Calling imputed data observed
Source circumstance: A statistical report inserts a model-based estimate for an incomplete response. Unsafe target: The target narrative calls the value directly measured.
Repair and diagnostic: Identify the imputation method and disclosure marker. An estimate’s uncertainty and provenance differ from an actual observation. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 18 — Converting JSON null to omitted property
Source circumstance: An API explicitly includes a field with null according to a contract. Unsafe target: Localization removes the field because no visible text exists.
Repair and diagnostic: Preserve the schema-level representation. API validation and consumer behaviour may distinguish an included null from an absent member. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 19 — Converting omitted property to null
Source circumstance: A message format uses omission to mean no instruction to change a field. Unsafe target: A translated integration guide inserts that field with null.
Repair and diagnostic: Do not infer null. In some documented systems that can request clearing a value rather than leaving it untouched. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 20 — Conflating empty string with null
Source circumstance: A user deliberately enters empty text, which the application stores distinctly from null. Unsafe target: A localized exporter normalizes it to null without a conversion rule.
Repair and diagnostic: Respect the defined data model. Empty text can be a valid string while null may mean no value; exceptions require explicit contract. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 21 — Treating NaN as unknown to a human
Source circumstance: A computation produces an undefined numerical result. Unsafe target: An interface translates it as user does not know.
Repair and diagnostic: Report a calculation error or undefined-result message appropriate to the system. The source of missingness is computational, not personal uncertainty. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 22 — Equating unavailable with intentionally withheld
Source circumstance: A remote server cannot be reached. Unsafe target: The target message says the organization is concealing the information.
Repair and diagnostic: Translate temporary access failure without inventing intent. An allegation of withholding changes the perceived cause and responsibility. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 23 — Equating withheld with not known
Source circumstance: A document deliberately protects information under an access rule. Unsafe target: The translated caption says authors do not know the information.
Repair and diagnostic: Keep the disclosure boundary. The owner may know the protected value; falsely claiming ignorance obscures its privacy or legal status. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 24 — Calling unsupported not applicable
Source circumstance: An older application cannot transmit a newly added data field. Unsafe target: The translated migration note says the property does not apply to those records.
Repair and diagnostic: Preserve the capability limitation. The property might apply but be unrepresentable in that system version. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 25 — Calling transport failure no data
Source circumstance: A report cannot be loaded because a file transfer failed. Unsafe target: The target alert reports that there is no underlying information.
Repair and diagnostic: Distinguish retrieval failure from absence of records. Otherwise users may destroy a valid archive or repeat data entry unnecessarily. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 26 — Replacing a stale snapshot with current state
Source circumstance: A cache displays pending although the authoritative record was completed yesterday. Unsafe target: The translation leaves out the last-updated timestamp.
Repair and diagnostic: Preserve provenance and timestamp while requesting a refresh. A correct label at an old time can be misleading as a current claim. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 27 — Translating machine codes as interface copy
Source circumstance: A controlled export requires the literal token NASK. Unsafe target: An editor substitutes a localized phrase directly into the code column.
Repair and diagnostic: Keep the exact token and localize only its display label. Changing codes can break interoperability even when readers prefer the new wording. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 28 — Losing the unselected option
Source circumstance: A form has yes, no and not answered as three distinct states. Unsafe target: The translated form forces every blank response into no.
Repair and diagnostic: Provide an unselected state and validate the mapping. Silence is not consent, rejection or an observed negative. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 29 — Dropping units from a withheld value
Source circumstance: A measurement is withheld but the field’s declared unit remains metres. Unsafe target: The translation removes the entire column label and unit.
Repair and diagnostic: Keep non-sensitive schema details where permitted. Future legitimate entries need the unit and the reason the present value is absent. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 30 — Using an unqualified dash
Source circumstance: A printed table uses a dash for several different reasons. Unsafe target: The translator copies the dash into an interactive chart with no accessible explanation.
Repair and diagnostic: Resolve each dash against the legend and provide an accessible label. Punctuation alone is not a reliable data dictionary. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 31 — Translating silence as the answer no
Source circumstance: A question asks whether the respondent has permission and receives no answer. Unsafe target: The target summary says permission denied or permission absent.
Repair and diagnostic: Preserve unanswered and follow the authorization rule. An unknown answer must not be translated into the person’s actual decision. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 32 — Calculating a percentage from a missing denominator
Source circumstance: A numerator of zero is recorded, but the denominator is not available. Unsafe target: The translated report presents 0% as an observed rate.
Repair and diagnostic: Keep the rate unavailable until a valid denominator and calculation rule are supplied. Zero count alone does not support every derived statistic. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 33 — Overstating temporal absolutes
Source circumstance: A label says not yet published or not currently available. Unsafe target: The translation claims never published or permanently unavailable.
Repair and diagnostic: Retain time reference and limits of knowledge. An absolute statement requires different evidence and may become false after an update. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 34 — Misplacing a locale decimal point
Source circumstance: The displayed zero is a genuine numeric value with a specified numeric format. Unsafe target: The localization inserts symbols that turn the value into a nonnumeric field or a different quantity.
Repair and diagnostic: Apply locale number rules without converting value type. The problem is numeric rendering, not missingness; verify alongside the numeric-fidelity chapter. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 35 — Mistaking a zero-width space for an empty field
Source circumstance: A copied cell looks blank but contains an invisible character. Unsafe target: The translation import treats it as a legitimate nonempty answer.
Repair and diagnostic: Inspect character-level data, normalize only under an approved rule, and do not invent a semantic answer for an invisible formatting artifact. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 36 — Losing a corrected missing-value history
Source circumstance: A record was first unknown and then verified. Unsafe target: The published translation overwrites the first status without a revision note.
Repair and diagnostic: Preserve the correction trail when governance requires it. The earlier uncertainty and the later fact should not be presented as a single timeless assertion. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 37 — Translating unreported as uncollected
Source circumstance: A dataset was collected but omitted from a public summary. Unsafe target: The target text says the data was never gathered.
Repair and diagnostic: Preserve reporting scope. Unpublished data can exist; the translation should not make claims about collection that the source does not support. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 38 — Confusing zero availability with empty inventory data
Source circumstance: A shop shows zero available units after an inventory check. Unsafe target: A translator changes the value to unknown because a blank appears in a neighbouring supply field.
Repair and diagnostic: Keep availability and supply metadata separate. Zero stock can be known while replenishment date remains unknown. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
Failure 39 — Leaving a safety-critical unknown implicit
Source circumstance: An operational alert lists an unknown sensor state. Unsafe target: The localized warning removes the qualifier to save screen space.
Repair and diagnostic: Use a short but unambiguous unknown-state notice and retain the recommended escalation. Do not produce a false normal or abnormal reading. The bilingual reviewer should identify the exact source state, compare the translated reader action with the intended action, and document a test case that would catch the same defect in a later release. If the classification cannot be recovered from the document, escalate to the data owner instead of assigning a plausible-looking value.
7. Worked diagnostics: forms, surveys and ordinary decisions
In the next cases, a source situation is described first, then a tempting but unsafe localization choice, followed by a defensible correction and a reviewer exercise. The cases are deliberately varied: some involve the value, others the cause of its absence, the permitted disclosure, the temporal state or the form’s branching logic. Every source and organization is fictional. For actual translation into a specific target language, the reviewed rendering must be natural in that language while preserving the same evidence and functional distinctions.
Worked diagnostic 001 — Registration form: optional secondary contact
Source situation. A registration form asks for a secondary contact only when one is designated. A person deliberately selects I have no secondary contact. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The export treats a blank contact-name field as unknown, prompting an employee to search for an unnamed person. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. The form must show no contact designated, not missing contact details. The distinction is between an affirmative absence of a relationship and lack of knowledge about a relationship. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Test a participant who designated a contact but left the name blank. That case truly requires follow-up and must not be merged with the no-contact choice. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 002 — Survey branch: workplace size
Source situation. A respondent identifies themselves as not currently operating an enterprise. The survey therefore skips a question about employee headcount. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The dashboard labels the skipped headcount as refused, reducing the reported willingness-to-answer rate. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Map the branch to not applicable under the survey design. The skip is a consequence of applicability, not a failure to elicit a response. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Compare this record with one from an operating enterprise that declines to reveal staff count. Their missingness reasons and analytic populations must differ. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 003 — Customer support: no ticket found
Source situation. An agent searches ticket records under one email address and finds no matching case. The customer claims to have used a different address. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translated response says the customer never opened a ticket and therefore cannot receive help. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. State no matching ticket was found using the supplied information. That is a bounded search result and leaves room for an identifier check. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Ask a second reviewer what search was performed. If they infer an exhaustive search of all identities, the translation overstates the evidence. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 004 — Attendance: zero absences
Source situation. A class log has complete records for the week. The verified number of absent participants is zero. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The localized report omits the zero to make the summary cleaner, suggesting the log was incomplete. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Preserve the observed count and coverage period. The phrase no recorded absences can be acceptable only if it does not falsely imply events could not occur outside coverage. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Test the contrasting case in which the attendance sheet itself was never submitted. This should remain unavailable, not the same zero. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 005 — Consent survey: no selected option
Source situation. The respondent has been shown separate yes and no choices but submits the page without selecting either. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translated export resolves the untouched control to no, though the form never recorded a refusal. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Represent unanswered and check the validation rule. In a consent context, nonresponse may prevent processing, but it is not an affirmative no. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Run the full form from an untouched default state. The stored output must distinguish no choice from an explicit negative selection. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 006 — Research survey: don’t know
Source situation. A participant is asked for an estimate and selects do not know after reading the question. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translated dataset maps the answer to blank, hiding that the participant engaged and chose a recognised response. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Keep the coded don’t-know response as a substantive response-status category when the methodology defines it that way. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Compare with a participant who skipped the question accidentally; a missingness audit should be able to tell the two apart. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 007 — Volunteer portal: declined details
Source situation. A volunteer completes registration but uses prefer not to say for demographic information. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The target-language profile says data could not be obtained because the volunteer did not understand the question. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Translate the deliberate choice respectfully and without implying inability. Preserve the preference in future exports. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Review the help text, privacy statement and reporting categories together; a euphemism must not silently reclassify the response. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 008 — Application: pending reference check
Source situation. An application has completed identity review but awaits a reference verification. Its stage is pending. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The localized status banner says application rejected or not eligible, changing a stage into a decision. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Describe the particular pending step if the interface permits it, or retain a neutral pending state. Avoid promising the outcome. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Test the same application after acceptance, rejection and cancellation. Each status must have a separate translation and user action. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 009 — Questionnaire: not asked for minor
Source situation. A form’s branching logic bypasses an adult-only financial question for respondents outside its applicability rule. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translated report records the absence as no response, suggesting a compliance problem. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Use the authorized not-applicable or skipped-by-design status and protect any sensitive explanation from unnecessary disclosure. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Compare the codebook’s rule to a respondent within the applicable group who leaves the answer blank; the two must not share a missingness code. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 010 — Appointment portal: no slot available
Source situation. A scheduling interface reports zero available appointments in a selected window but can still access the schedule. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translation says appointment data unavailable, inviting the user to refresh rather than consider another window. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. State no appointments currently available for that specified window. Zero availability is a real computed state, not a system outage. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Disconnect the schedule server in a test environment. That state needs a different localized message and a different recovery action. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 011 — Delivery form: unknown arrival date
Source situation. A package has been dispatched but a reliable arrival date has not yet been determined. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The target message inserts tomorrow, relying on a usual delivery estimate not approved for this item. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. State estimated date unavailable or not yet confirmed, according to the system. A common expectation is not evidence of a promise. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. When a validated forecast arrives, verify that the translated page changes state without implying the earlier omission was a mistake. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 012 — Public consultation: unanswered question
Source situation. A consultation document lists a question the committee has not yet answered. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The localized summary prints no, implying a negative policy decision. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Preserve unresolved or unanswered. Readers must distinguish a pending policy answer from an adopted position. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Ask whether the committee has considered the issue or simply has not reached it. If that distinction is material, seek precise source wording. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 013 — Return form: not required
Source situation. A product-return form says a receipt is not required for the selected return method. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translation describes the receipt as missing, leading customers to supply unnecessary personal documents. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Preserve nonrequirement as a rule, and distinguish it from a required receipt that a customer cannot find. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Compare the options in the workflow; the visible text and validation logic should agree on which attachments are mandatory. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 014 — Travel survey: no response versus zero trips
Source situation. One respondent states zero trips during the reference month. A second respondent does not complete the travel section. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translated dashboard combines both as none, reducing coverage and biasing apparent trip frequency. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Keep zero trips as a legitimate answer and nonresponse as a missing-data status. The same number of blank-looking interface cells can have different analytic meaning. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Check numerator, denominator and response-rate calculations in the published chart after localization. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 015 — Public archive: closed item
Source situation. A finding aid contains a record whose detailed description is restricted for a period. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translation calls the record absent from the archive, obscuring the distinction between existence and access. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Use a disclosure-appropriate restricted description without revealing withheld information. The archive may retain a visible identifier and access policy. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Test with an entirely unaccessioned record; do not assign restricted status to a record that simply does not exist in the catalogue. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 016 — Event RSVP: no reply
Source situation. An invitation has been delivered, and the recipient has not responded. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The localized organizer dashboard says declined, turning silence into an affirmative choice. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Translate no reply received. A reminder may be appropriate according to policy; a decline should only follow an explicit response. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Compare declined, accepted, maybe and not responded in the entire workflow. Each must survive export and aggregation. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 017 — Profile field: system-default language
Source situation. A user account has a language preference set by automatic detection. The user has never selected it manually. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. A translated profile says preferred language chosen by the user, misleading a support agent about the preference’s origin. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Show automatically selected language or distinguish user-set from system-inferred values in the log if the interface exposes provenance. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Change the user selection deliberately and recheck. A later explicit choice should replace the default provenance without erasing historical events where relevant. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 018 — Exam records: result not issued
Source situation. A result is awaiting final quality review and has not yet been released to the candidate. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The localised transcript inserts a failing grade to fill the empty mark field. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Keep not released or pending result, following the authorized examination code. No academic judgement can be inferred. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Test the final released record and any withdrawn result separately. Do not silently change a data state into a score. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 019 — Feedback survey: questionnaire not completed
Source situation. A customer begins but abandons a questionnaire after two items; the remaining questions were never displayed. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translation classifies the remaining fields as refused, claiming choices that were never offered. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Record incomplete survey or not presented, depending on the study’s codebook. Preserve the sequence that actually occurred. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Simulate completing the form up to different points and confirm that each unasked item remains distinct from an explicit decline. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
Worked diagnostic 020 — Annual report: figure not reported
Source situation. A report includes a column for a statistic but leaves the year blank because the figure was not published in that edition. A translator must identify both what the field appears to show and what evidence its workflow actually supplies; those are not always the same thing.
Tempting but unsafe version. The translation announces that the organization did not collect any such data. The risk is an altered conclusion or action, not merely an awkward word. A polished sentence is inadequate if it changes the data state.
Safer rendering and why. Use not reported or figure unavailable in this edition, and avoid describing unseen source processes. The choice should be checked against the form logic, glossary and field definition, then phrased naturally in the target language without adding a claim that the source never made.
Independent diagnostic. Look for a companion methodological note. If no explanation exists, a transparent unresolved status is safer than a guess. A reviewer should write down the expected state and test a counterexample, rather than accepting the translation because the sentence resembles familiar UI copy.
8. Worked diagnostics: software, data exchange, research, reporting and high-stakes records
These technical cases add a second layer to the earlier human-form examples. They show why user-facing text and stored state must be reviewed together. The samples are conceptual descriptions rather than runnable code or specifications for any actual organization. Where a formal data contract exists, its documented version prevails; where a legal, medical or safety classification governs a decision, qualified review is essential. In every case, the translation task is to preserve meaning, not to silently repair a database.
Worked diagnostic 021 — CSV import: an unannotated empty cell
Source and context. A CSV column contains an empty cell while the project dictionary says blanks are undecided, not zero. Some software tools would automatically import the cell as null. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. A translation vendor formats the empty cell as the numeral 0 because the column is numeric, generating an unsupported observation. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep the cell’s documented undecided state and confirm the import dialect, column metadata and missingness convention before transforming the file. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Reimport round-trip data and compare the record identifier, original cell, semantic value and derived summary; do not equate a file parsing default with the author’s intent. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 022 — JSON API: member omitted on purpose
Source and context. An update endpoint’s own contract defines omitted fields as unchanged. A partial payload contains no address member, despite a prior address existing. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The target-language integration example includes address: null, claiming to clarify the missing data but potentially instructing the server differently. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Describe omission as no change to the field under that particular contract. Preserve examples of omitted members exactly instead of inserting prose-oriented placeholders. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Create three separate tests for omitted, explicit null and an explicit empty string; confirm behavior against the actual endpoint contract rather than a generic API assumption. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 023 — JSON API: explicit null recorded
Source and context. A record contains an explicit JSON null in a field whose schema permits it, while another field is absent entirely from the object. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The localized documentation says both fields are empty, making two technical states appear interchangeable to an integrator. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep null as the technical token in code and explain its project-specific semantic interpretation separately in translated documentation. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare validation and output serialization for each case, remembering that JSON null has a defined type but not one universal application meaning. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 024 — Empty array: zero members is not no value
Source and context. A service returns an empty array of active appointments after a valid query. The application distinguishes this successful response from a failed query. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated alert says appointments information could not be retrieved, converting a known empty result into a technical outage. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Render no matching appointments for that query, while preserving the fact that the request was fulfilled. An empty collection need not mean missing collection. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Simulate an HTTP error, a timed-out request and a successful empty list. The local-language UI must not merge their recovery actions. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 025 — Empty string: deliberate input
Source and context. A profile allows the user to remove an optional tagline by saving an empty text value. The system distinguishes that from no submitted change. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated audit message says tagline unknown, suggesting missing information rather than a deliberate clearing operation. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve the user’s action and its new display state. If the operation is described as cleared, that must be justified by the application’s semantics. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test the three states: never specified, explicitly cleared and unchanged in a partial update. Translate the status and the event separately. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 026 — Sentinel code: minus 999 is not a quantity
Source and context. A legacy dataset uses -999 as a documented sentinel for missing observations in a temperature column, not as a measured temperature. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The target report converts the sentinel to a real negative temperature and includes it in the average. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Map the sentinel according to the data dictionary before calculations or locale formatting. A translation can explain the sentinel, but it must not create a measurement. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Inspect the import script and recompute summary statistics after excluding code values as directed. Never invent a replacement numerical reading. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 027 — NaN: computation rather than refusal
Source and context. A calculation step produces NaN after a division operation is undefined under the software’s numeric rules. No respondent supplied an answer. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated dashboard labels the field respondent declined, inaccurately blaming a person for a computational condition. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Classify the source as undefined numeric computation or a processing error as appropriate to the product, without exposing raw jargon to an unsuitable audience. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Repeat with valid inputs, zero denominators and missing operands; ensure the source of the undefined result remains distinguishable in logs. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 028 — Zero baseline: actual comparison exists
Source and context. A laboratory’s comparison dataset records a baseline of zero for an instrument under a defined calibration procedure. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated chart treats that zero as no baseline, suppressing a valid comparison point and making later change impossible to interpret. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve a legitimate numeric zero. Whether the method makes that baseline scientifically meaningful is a separate domain question. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Inspect instrument metadata, calibration notes and units. Compare the chart before and after translation for omitted bars and altered axes. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 029 — Percentage: missing denominator
Source and context. A report contains a documented count of incidents but lacks the eligible population denominator for the reporting interval. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The localization generates 0% because the percentage widget defaults to zero when its denominator field is blank. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. State percentage unavailable or not calculable under the documented method; retain the known incident count without inventing a rate. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Verify denominator provenance, time window and any exclusion criteria before displaying a derived percentage. The absence of a rate is not proof of absence of incidents. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 030 — Chart: interrupted data collection
Source and context. A time-series graph has no value during one interval because the collection unit was offline; surrounding intervals have valid readings. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The localized graphic connects adjacent points and describes the interval as stable, inventing continuity across missing observations. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep a visible gap or documented missing-state marker. Do not interpolate unless the source explicitly supplies an estimation method and the estimate is labelled. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Ask whether a chart-reader could distinguish an observed plateau from a connection drawn through missing time. Test accessible descriptions, too. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 031 — Default: unselected options in a form
Source and context. A purchasing form contains a default delivery preference assigned by the system, and the user has not confirmed it. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated receipt states the customer selected express delivery, making a system default look like a deliberate instruction. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Retain the default’s provenance and distinguish whether it counts as a confirmed selection under the actual ordering rule. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Inspect user action logs. Compare the initial default state with a later explicit choice and with a request that makes no delivery selection. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 032 — Unsupported: old client version
Source and context. A newer export includes an optional field that an older client cannot display, although the underlying property may exist for its records. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated manual calls the field not applicable for old clients, implying a property-level difference between users. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. State not supported by this client version, and indicate supported migration or update paths only where documented. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare the same record viewed through old and new clients. If the value appears in one but not the other, inapplicability is the wrong explanation. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 033 — Feature disabled by configuration
Source and context. A settings panel displays no analytics results because the analytics feature is disabled for the workspace. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The local-language message says no activity occurred during the month, fabricating a zero-activity finding. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Translate disabled or data unavailable due to configuration as appropriate, and avoid inferring metrics for a period without tracking. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Enable the feature in a controlled test and verify whether new activity is collected prospectively or any historical data can legitimately appear. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 034 — Service temporarily rate-limited
Source and context. A request encounters a documented request limit, so the application cannot display new results until a later attempt. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation displays no items found and encourages unnecessary data re-entry instead of retrying after the stated condition. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Explain that retrieval is temporarily limited. Preserve an accurate retry instruction if the product provides one; do not invent exact restoration times. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Check whether the status differs from authentication failure, network outage, completed empty query and permanent access restriction. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 035 — Access denied: known record protected
Source and context. A user lacks permission to read an existing record. The system responds with a controlled access message that reveals no protected details. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated interface says the record does not exist, potentially concealing an important authorization problem or violating the intended disclosure design. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Use the security team’s approved user-facing message. Some systems intentionally mask existence, so translation must preserve the authorized wording and not reveal more. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test under accounts with different permissions and compare messages with the approved disclosure policy. Never guess the protected value. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 036 — Redacted diagnostic log
Source and context. A published incident report replaces personal identifiers with redaction markers but keeps timestamps and technical events visible. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation renders the markers as deleted data, falsely suggesting the original event log had been altered or destroyed. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep redacted or masked according to the release process. The human explanation should differentiate an edited public view from the underlying retained record. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare disclosed text with the publication policy. A translator must not reconstruct, speculate about or reveal removed personal details. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 037 — Contract table: empty signature date
Source and context. A contract has an unfilled date field next to a signature line, and the document does not establish whether signing occurred. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated note says the party never signed, converting a blank printed field into a historical conclusion. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Describe the field as not completed in the available copy, leaving the signing event unresolved unless supported by other evidence. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Ask which authoritative document version was supplied. A second signed copy may exist; a blank visible field cannot settle its existence. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 038 — Data migration: unknown legacy field
Source and context. A migration team cannot determine the meaning of a legacy code in a retired database. Records still contain that code. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated mapping guide assigns it a popular modern category to make all labels readable. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Record unmapped legacy code and the unresolved interpretation. Guessing a likely category can corrupt migrated records and mislead users. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Retain raw values, provenance and migration warnings. Ask the source owner for archival documentation before establishing an equivalence table. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 039 — Schema version: changed absence semantics
Source and context. A new schema version changes how certain optional fields are represented while historical records preserve older conventions. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation applies the newest meanings retroactively to every row, erasing a version-specific distinction. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Separate the versions and label each representation according to its governing schema. A stable display label must not conceal incompatible underlying states. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Sample records from before and after the migration. Verify both source versions and any documented conversion rules without rewriting history. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 040 — Placeholder: variable looks like text
Source and context. A message template displays a localized label next to a placeholder that will later carry a dynamic value. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. An editor translates the placeholder identifier or inserts a literal zero because the preview shows no sample value. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve the placeholder contract and localize the surrounding words. Preview emptiness does not license the creation of a fabricated data value. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Run renders with zero, null, unknown, long localized text and special characters, checking both grammar and markup. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 041 — Controlled missingness code in a CSV
Source and context. A public dataset defines code NASK as the question not being asked under survey logic. Its code values must remain stable. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translator replaces NASK with a translated full phrase inside the machine-readable export, breaking existing consumers. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Translate the codebook display label, not the data code. If a localized CSV is genuinely required, provide a documented mapped presentation column. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Validate row counts, codes and joins against the original dataset. The round trip must not turn a controlled absence reason into free text. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 042 — Boolean false: observed decision is not missing
Source and context. An API includes a boolean false for a completed negative answer to a yes/no check. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The target-language report hides false values because the rendering library treats them as empty, making completed checks seem unperformed. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Restore an explicit no or negative label matching the field’s definition. False is a genuine boolean value, not an absent record. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test true, false, null and omitted separately, including keyboard and screen-reader output for every state. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 043 — SQL null: query predicate uncertainty
Source and context. A database field contains SQL NULL and a report query uses comparison logic under SQL’s three-valued semantics. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation guide describes NULL as the literal number zero, leading readers to expect ordinary arithmetic comparisons. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Explain NULL according to the database and query semantics. It is not a conventional number or text string to be globally substituted. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Have an engineer test conditions, aggregations and null-sensitive predicates. Translators should preserve conceptual distinction without prescribing unverified SQL behavior. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 044 — Cached results: old unknown state
Source and context. A report was pending when a cached snapshot was generated, but the authoritative source may since have updated. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation removes the snapshot timestamp and presents pending as the unequivocal current status. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Retain the as-of time and refresh instruction where documented. A stale state may have been accurate historically but unreliable now. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare the cached display with its source record and update metadata; do not quietly pretend an older state was always current. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 045 — Telemetry: silent equipment
Source and context. A monitoring dashboard receives no messages from a sensor for a period. It cannot determine the actual operating temperature. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The target warning reads safe temperature zero, an invented measurement with a potentially dangerous consequence. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. State telemetry unavailable or sensor not responding. In safety contexts use an approved escalation message, not a guessed numeric substitute. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Disconnect the sensor intentionally in a controlled test. Compare its display with a live sensor legitimately reporting zero in the relevant unit. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 046 — Measurement: below a laboratory detection limit
Source and context. An analytical report records a response below the specified method detection threshold but confirms a measurement procedure was performed. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation converts the result into none detected as an absolute factual assertion, eliminating the method boundary. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Retain below method detection limit, including the threshold and units where supplied; domain specialists should approve reporting conventions. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Check how the method’s blank, invalid run, genuine zero and below-limit categories differ. Do not merge them for brevity. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 047 — Quality control: rejected sample
Source and context. A sensor produced a numeric reading, but quality control rejected the result because calibration criteria were not met. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated graph publishes the number as validated and excludes the rejection marker. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve the rejected or invalid quality flag. A value can be numerically present yet not acceptable for use. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Inspect source QC metadata and downstream calculations. A rejected value should not enter validated totals unless the approved method explicitly permits it. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 048 — Audit: finding not examined
Source and context. An auditor’s checklist states that a particular control was not tested during this engagement. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation says the control passed because no adverse observation was recorded. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep not evaluated or out of audit scope according to the report. The absence of a finding cannot prove a control’s compliance. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare the audit plan, tests actually performed and final opinion boundaries. Do not let a local-language shorthand broaden assurance. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 049 — Inventory: zero stock versus missing reconciliation
Source and context. The warehouse count confirms zero units of one item, while another item’s physical count has not yet been reconciled. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated stock page marks both as out of stock, potentially discouraging valid orders for the unreconciled item. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve available quantity zero only where verified. Label unreconciled stock with the documented state and explain any ordering restriction separately. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test purchasing, stock transfer and stocktake reports across confirmed zero, positive stock and unknown availability. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 050 — Budget: no allocation versus no report
Source and context. One project has an explicitly approved allocation of zero, while another has not submitted its planned budget. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated management table shows zero for both and silently treats non-submission as a financial decision. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep approved zero distinct from budget not submitted. Their governance consequences, approvals and follow-up actions differ. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Ask which evidence supports approval and whether the reporting period is complete. A missing budget document should not be called a vote against funding. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 051 — Payment ledger: no payment recorded
Source and context. A portal’s ledger shows no received payment by a particular cutoff, but funds might still be in transfer. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated receipt declares the payer never paid, a broader historical accusation. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. State no payment recorded as of the cutoff and follow the authorized reconciliation process. The ledger and the payer’s action are different evidence sets. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare paid-and-cleared, initiated, pending, failed and not recorded; ensure receipt messages reflect the chosen event model. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 052 — Banking interface: value masked
Source and context. A financial interface masks most digits of a stored account number to protect the user’s information. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation says account number missing and prompts the user to enter a replacement in an unrelated field. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep the masking explanation and reveal only the authorized subset. Do not imply a record has been lost. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Review the interface under authorized and unauthorized roles and check that translations do not expose hidden digits through labels or alternate text. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 053 — Contract price: to be determined
Source and context. A draft agreement specifies that a price will be determined by a later documented procedure. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated schedule fills zero, making a provisional pricing gap appear to be a free supply obligation. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve price to be determined and the condition or process if stated. A future determination is neither a zero price nor necessarily unknown forever. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare draft, signed and amended versions. If a final figure appears later, keep the timing and authority of each revision. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 054 — Sales page: free versus price unavailable
Source and context. A product is advertised as free in one campaign; a different product has no published price until a quotation is prepared. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation uses free of charge for both because a missing price is displayed as a dash. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Separate zero price under an offer from undisclosed or quotation-based pricing. The user’s purchase expectations are legally and commercially different. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Inspect checkout logic and offer conditions. A missing price cannot justify creating an enforceable-looking free offer. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 055 — Warranty: field not applicable to service
Source and context. A service plan lacks a hardware-serial-number field because it does not cover a physical device. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated service dashboard says serial number not found, leading staff to search for nonexistent equipment. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Mark the field inapplicable to this product type under the service taxonomy. Do not classify the record as incomplete. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test a device-based product missing a required serial number. That is a different error path and must remain visible. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 056 — Health record: unknown versus declined
Source and context. A clinical data-entry screen distinguishes the patient not knowing a detail from the patient declining to disclose it. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation uses unknown for both, obscuring the collection history and potentially affecting subsequent questions. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Follow the locally authorized clinical terminology and the specific system’s code definition. Preserve dignity and privacy without inventing a medical history. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Have qualified clinical and language reviewers test the labels with real workflow logic; this example is about record states, not diagnosis or treatment. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 057 — Health record: observation not performed
Source and context. An ordered observation was not carried out, so there is no measured value. The order and cancellation records remain visible. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated report displays normal, treating the lack of an abnormal result as a positive clinical finding. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve test not performed or the precise workflow reason authorized by the system. There is no result to classify as normal or abnormal. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Inspect whether cancellation, refusal, unavailable sample and completed negative result each have distinct codes and messages. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 058 — Publication table: privacy suppression
Source and context. A research appendix hides a very small category cell while disclosing adjacent larger counts, using a specified suppression symbol. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translator writes 0 or guesses a small value from nearby totals, risking erroneous inference and possibly defeating confidentiality controls. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Retain the approved suppression sign and explanation. Do not reverse-engineer hidden figures into translated publications. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Review secondary disclosure risks with the data steward. A value protected by the source must remain protected after translation and reformatting. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 059 — Research dataset: adjusted estimate
Source and context. An analysis has incomplete observations and reports a model-based estimate flagged as imputed. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation calls the value measured after replacing missing observations, hiding the analytical assumption from readers. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Keep the method flag, cite the published analysis procedure where available and maintain the distinction between observed and estimated records. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Ask whether a user can reproduce which entries were imputed. Test that translated legend and footnotes continue to disclose imputation. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 060 — Environmental record: station offline
Source and context. An environmental station shows no value for a day because power was unavailable, not because pollution was absent. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated summary says zero pollution and draws the data line at zero for that day. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Represent an interruption in data acquisition. If an estimated value is later added, it needs its own provenance and qualifier. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare data coverage and instruments before translating a daily status into an environmental conclusion. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 061 — Legal file: withheld document attachment
Source and context. A public docket lists an attachment but its contents are not available for public access. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation states no attachment exists, erasing the difference between a filed item and its disclosure state. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Use the approved restricted-attachment label, accurately conveying only the publicly authorized facts. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Check with the court or records authority’s official display rules; do not infer the hidden attachment’s content. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 062 — Education record: mark awaiting release
Source and context. An assessment has been completed, but results have not yet been issued because formal moderation is still in progress. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated portal labels the missing grade fail, implying a substantive assessment result. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve awaiting release. The eventual grade must be provided only when authorized and confirmed. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test withheld, pending, cancelled and released outcome states independently, preserving the distinction between an administrative step and a judgement. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 063 — Research catalogue: unknown creator
Source and context. A museum or archive describes an object whose creator has not been established by available evidence. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation invents anonymous as a definitive artistic attribution or says no creator existed. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Use unknown creator or attribution undetermined as supported by the catalogue, distinguishing missing knowledge from an intentional anonymous work. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Review provenance notes and earlier disputed attributions; do not turn scholarly uncertainty into categorical identity. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 064 — Transit timetable: journey not scheduled
Source and context. A schedule contains no service at a particular requested time because the service does not operate in that window. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation says timetable unavailable, implying a technical failure rather than a scheduling fact. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. State no scheduled service for the relevant interval, retaining route and date constraints. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare public holidays, temporary suspension and unavailable live data as different status sources; never invent a schedule from missing information. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 065 — Digital certificate: revocation status unknown
Source and context. A verification client cannot reach an authority to retrieve a certificate’s status. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated banner says certificate valid, interpreting the lack of a revocation warning as positive proof. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. State verification could not be completed and follow the application’s established trust policy. Unavailable evidence is not proof of validity. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Test signed, expired, revoked and status-unavailable states, keeping the security consequence separate from any localized aesthetic choices. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 066 — Machine translation: hallucinated missing field
Source and context. A machine-translated report contains a plausible target sentence filling a field the source left unresolved. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. A proofreader accepts the fluent sentence without checking source data, allowing a fabricated figure or outcome to appear authoritative. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Compare every data-bearing claim to the source and recorded field state. A reviewer should reject factual additions, not merely fix their grammar. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Run source-target alignment and a back-to-source challenge for absence terms; any numerical value not supported by input requires explicit verification. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 067 — Document conversion: lost footnote
Source and context. A document conversion preserves the main table but drops a footnote explaining that a dash means not collected. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated PDF presents ambiguous dashes with no legend, making missingness impossible to interpret correctly. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Restore the footnote or an equally accessible reference. Preserve the approved status legend alongside the translated table. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Review the rendered output after conversion and verify that markers, legends and table associations survive screen reading and export. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 068 — Identity form: not yet verified
Source and context. An identity detail is supplied by a user but remains unverified against the approved source. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated confirmation says verified, raising the claim’s evidential status beyond what the source permits. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Use submitted or verification pending as defined by the workflow; do not erase the fact that a claim exists. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare pending, verified, rejected and not supplied. Keep the field value’s existence separate from assurance about its correctness. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 069 — Migration report: legacy data lost
Source and context. An archive migration log states that several original files became unavailable after a failed process, with an unresolved recovery investigation. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translation labels files never collected, concealing the known historical presence and the migration failure. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Report data loss or temporarily inaccessible material as established by evidence and preserve the investigation’s status. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Check the original inventory and recovery log. Do not infer permanent loss while restoration remains possible. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 070 — Food-label database: allergen field blank
Source and context. A product database lacks complete source documentation for an allergen field; there is no reviewed statement that the substance is absent. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The localization renders the field allergen free, treating missing evidence as a safety assurance. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Use an authorized information-unavailable warning and escalate for specialist review. Never convert silence into a safety-related claim. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare fully reviewed negative, positive and undocumented statuses. Publication should respect regulatory and professional controls. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 071 — Insurance claim: coverage not assessed
Source and context. A claim has been received but a coverage determination has not yet been made. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The translated portal says claim uncovered, reversing an undecided state into a contractual decision. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve not yet determined and any documented next step without promising eligibility or payment. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Check whether the claim is submitted, under review, approved, denied or awaiting documents. Each status supports different customer actions. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
Worked diagnostic 072 — Manufacturing report: measurement invalid
Source and context. A production instrument outputs a number but a control flag marks the run as invalid after a calibration failure. The evidence includes a field, its definition and a surrounding workflow; translating only the cell’s visible characters would be insufficient.
Failure pathway. The localized quality sheet keeps the number but drops its invalidity marker, presenting it as a released measurement. The error is substantive because it changes what a downstream person, calculation or validator believes happened.
Faithful translation response. Preserve invalid result and the reason if authorized, rather than substituting zero or deleting the evidence trail. The target should remain appropriately concise for its channel while retaining provenance, scope and the relevant boundary of knowledge.
Adversarial review test. Compare instrument logs, traceable measurement records and release approvals. A value’s visual presence does not mean it may be used. A second reviewer should test a contrasting state and ensure that the two localized outcomes do not collapse to the same claim or action.
9. Transfer laboratory: apply the method to new source languages, domains and channels
A translator who can distinguish missingness in an English gloss has not yet demonstrated equivalent competence in a live language pair. Another language may use the same everyday word for several absent-data situations, or may force a distinction English leaves implicit. The exercise is therefore to identify the meaning first, then choose an attested, natural target-language expression, checking the data code and the intended reader action. Do not translate these examples mechanically; each asks for a domain decision before lexical choice.
Transfer workshop 01 — Two incident rows
Source exercise: The source table reads Week A: zero incidents verified; Week B: no data due to reporting outage. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: A safe translation retains the verified zero and the separate outage. The reader must not assume that Week B also had zero incidents. If the target language uses one shorthand marker for both, write a legend that restores the distinction. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Ask a reviewer to identify which week supports a numerical claim. A translation fails when both look like completed negative observations. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 02 — Three survey outcomes
Source exercise: A source questionnaire contains: did not know, declined to answer, and question not shown because of branching. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Choose three different response-status categories even if all could appear blank in a naive export. The first concerns knowledge, the second willingness, and the third the design of collection. Respect respondent dignity. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Have a survey analyst compute response-status frequencies from the localized labels. The categories must reproduce the source distinctions. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 03 — When N/A means no application
Source exercise: A printed report shows N/A in a column without a legend, and the compiler has not specified whether it means not applicable or unavailable. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Do not expand the abbreviation by guesswork. Keep a neutral placeholder temporarily and request the codebook or original author’s clarification. A single abbreviation can encode different states in different projects. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Test whether two reviewers independently assign the same meaning. Disagreement is a reason to query, not a license to choose the more attractive phrase. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 04 — Masked account digits
Source exercise: A screen contains an account number formatted as several mask characters followed by the last four digits. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: The code and visual pattern must follow the system’s privacy rules. The translated label can indicate a hidden number without claiming the account is missing or inviting the user to disclose it. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Check screenshots, screen readers and copy actions under authorized roles. A correct translation must not inadvertently reveal hidden content. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 05 — Test not performed
Source exercise: A specimen was not tested because an approved workflow halted before measurement. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Translate the procedure state, not a clinical conclusion. A statement that the result was normal, negative, or zero would invent a test outcome. Use the source organization’s qualified terminology. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Show the target to an appropriate specialist who can contrast canceled, performed-negative and performed-positive events. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 06 — Omitted JSON member
Source exercise: A technical reference describes a payload in which omitting the property requests no update; explicit null has a different specified effect. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Keep the example syntax and explain the difference in natural target language. Do not add an attribute for readability or equate null with omission. The scope is this API’s contract, not all APIs. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Validate examples against the published contract and use automated tests for both conditions. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 07 — Zero in pluralized copy
Source exercise: A product shows the count 0 beside a localized message about available reservations. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Use the correct locale’s grammatical rules for the numeral and a natural zero-message if the product requires one. But preserve zero as a known availability count; do not recode it as not loaded. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Test 0, 1, 2 and other language-relevant quantities, plus a separate server-error state. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 08 — Skipped financial question
Source exercise: A form skips a financial field for a category of applicant for whom the question does not apply under written instructions. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: The translated form should preserve the scope condition and explain why input is unnecessary only to the extent approved. It must not describe the applicant as refusing to cooperate. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Compare applicant categories against actual form branching and validation rules rather than guessing from a sample screenshot. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 09 — Missing denominator in a percentage
Source exercise: A graphic lists 0 cases but the relevant population count is unrecorded. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: The count may remain zero, while the rate is unavailable. A mathematical percentage requires a defined denominator and method, and the source did not supply them. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Have the reviewer check whether the target graph falsely presents 0% or a safe rate. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 10 — Bounced database query
Source exercise: A user searches for a record; the database returns a timeout, not a completed no-match result. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: The target message must signal that the search was not completed. No records found would be an ungrounded conclusion. Retry instructions should reflect the actual product response. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Contrast completed empty search, invalid credentials and timeout in the interface. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 11 — Confidentiality-suppressed cell
Source exercise: A table shows a controlled symbol indicating a deliberately suppressed count. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Keep the suppression marker and explanatory legend, or apply an approved localized display equivalent without changing the protected state. Avoid calculating hidden values from adjacent published totals. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Review the published table’s disclosure risks and its accessible alternatives. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 12 — Unresolved template placeholder
Source exercise: A translator sees the string Orders: {count} but the preview displays no sample number. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Translate the grammatical surrounding phrase without altering {count}; the variable is not a missing observed value. Render valid count values and missingness states separately. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Test interpolation across target-language word order and plural branches. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 13 — Old and new schemas
Source exercise: A system has two documented versions: the old version uses a sentinel; the new one has an explicit missing-state field. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Translate both conventions by version and clearly distinguish the migration rule from mere wording changes. A conversion is a data operation needing authorization, not a synonym replacement. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Ask a technical owner to identify the source version for each record. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 14 — Redacted public attachment
Source exercise: A public document contains passages withheld from the released copy while a private original is controlled. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Describe redacted material according to the disclosure policy; do not assert that nothing was originally written. Do not infer the missing information from contextual clues. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Check that translated text does not broaden access or disclose restricted identities. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 15 — Gap in a daily time series
Source exercise: A graph loses one day of readings because the recorder stopped, while values before and after are available. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Keep the unavailable interval visible in captions and accessible descriptions. Any line drawn across it is a presentation choice, not proof that the source measured intermediate values. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Contrast missing period, exact zero, estimate and interpolation in the final graphic. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 16 — Estimated replacement for missing responses
Source exercise: A statistical appendix indicates that some observations were imputed under a specified model. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Preserve the imputation qualifier and do not imply direct observation. The translation may explain imputed in reader-friendly prose, while the protected methodological definition remains the authority. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Confirm that methods, tables and footnotes consistently mark derived records. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 17 — Pending decision in a portal
Source exercise: A claim has been submitted and remains under review, with no eligibility determination. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Use pending or under review as defined by the institution. Neither approved nor denied can be inferred; a status message should not promise a decision time without evidence. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Compare approved, denied, incomplete and under review in target-language usability tests. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 18 — Below-limit laboratory reading
Source exercise: A report says the concentration was below the method’s detection threshold, while the instrument run was valid. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: The statement does not justify zero concentration. Translate the detection relationship and units faithfully, and let qualified specialists confirm domain conventions. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Contrast below limit, not tested, invalid test and observed numerical result. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 19 — Unanswered invitation
Source exercise: A person received an invitation but did not respond by the date of the report. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Translate no response received by the stated date, not declined. A nonresponse may cause a workflow action under local policy, but it is not a record of the recipient’s decision. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Ask a recipient whether the message attributes a decision they never communicated. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 20 — Consent selection left untouched
Source exercise: A consent interface includes an unselected state in addition to yes and no. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Preserve the absence of an affirmative or negative response. The application may deny access until a valid choice is made, but that rule is separate from translating the user’s answer. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Test screen-reader announcements and stored values for untouched, yes and no. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 21 — Fabricated machine-translation number
Source exercise: An automated draft inserts a plausible count into a table that had an unreported value. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Reject the invented count, restore the source’s actual status and record an accuracy issue. Fluency or common sense cannot substitute for documentary evidence. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Challenge the draft to identify its exact source cell and provenance; unsupported specificity fails. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 22 — Corrected data state
Source exercise: An audit initially marked a field unknown and later verified a definite value. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: A faithful history differentiates the first uncertainty from the later correction, including the effective time where required. It should not imply the earlier editor knew the later answer. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Compare the source versions and translated changelog for lost chronology. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 23 — Accessible table dashes
Source exercise: A report uses two different dashes with distinct legend meanings. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Preserve their distinctions in text alternatives and visual presentation. A screen-reader user should not hear identical unexplained punctuation for states requiring different interpretation. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Check rendered PDF, HTML table headers, cell descriptions and downloaded CSV. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 24 — A preference assigned automatically
Source exercise: A customer profile shows a default setting inferred by the system; the customer has not expressly chosen it. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Translate the setting and its provenance separately. Do not present the default as a deliberate user preference, especially in a support history or consent-related context. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Test later explicit selection and confirm that the user-selected state is now labelled differently. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
Transfer workshop 25 — A search with no matching source
Source exercise: An archive search returns no match under one spelling, while an alternative transliteration could identify the item. Before drafting, write a one-line semantic contract: value existence; knowledge owner; collection status; disclosure status; representation; time; next decision.
Model decision: Translate the bounded search result and avoid claiming that no historical record exists. Search conditions and name variants are part of the evidence. The model answers describe the decisions to protect, not mandatory literal translations into all languages.
Pass/fail diagnostic: Try a second query and verify that the wording remains defensible whichever result appears. The pass criterion is the same permissible inference and next action, supported by the same source evidence. Record a revised target rendering if the first draft fails.
10. The editorial warehouse: a practical ledger, acceptance tests and release checklist
A reusable field ledger
Build a table with these columns for every data-bearing source: record identifier; source field; source value; source token or code; data type; source status label; absence reason if known; governing version; source confidence; collection event; disclosure rule; time of assertion; target display label; target explanation; target-language reviewer; unresolved question; expected computation; permitted reader action; and test result. Keep the technical code unchanged unless the actual data-migration requirement authorizes a mapped code. This separation prevents a well-intentioned translation from becoming an unauthorized database transformation.
A minimal set of acceptance tests
For a field that can contain numbers, include tests with genuine zero, positive quantity, negative quantity if valid, null, absent property, empty text if valid, not applicable, unknown, user-declined, withheld and error. Some fields cannot have every state; document that as part of the schema rather than manufacturing synthetic invalid cases. For a form, test whether the participant saw the question, answered it, declined it or was prevented by branching. For a report, test whether the data was gathered, accepted, published and accessible. For a security-restricted record, test that a translated screen reveals no more than the approved public version.
Severity and release priority
A typographic dash problem may be low severity in a decorative mock-up, but a dash that converts an unknown medical observation into a normal result or turns a hidden account number into an invitation to disclose private data is much more serious. Prioritize by probable reader action and harm, not by word count. Record whether the defect is an accuracy distortion, unapproved addition, omission, terminology mismatch, locale-format problem or accessibility failure. The MQM framework can help organize this assessment, but the project’s risk policy determines appropriate release gates.
A bilingual review conversation
Ask the source owner: What does this empty-looking field mean here? What evidence determines its status? Who knows the value? Was the question asked or procedure performed? Is the field inapplicable or merely uncollected? Is the value restricted? What is the permitted next step? Ask the target reviewer: Does this phrase imply zero, absence, ignorance, refusal or deliberate withholding? Does it sound accusatory? Does its grammar imply a completed event? Can a reader distinguish pending from a final decision? The best answers come from comparing real workflow states, not debating isolated synonyms.
Machine-translation evaluation without invented confidence
Do not assume an AI-translated or automatically localized label is correct because it looks plausible. Use an anchored source-to-target checklist. Extract every data-bearing phrase, code, number, unit, status, negation, person and time qualifier. Compare each item against the source. Ask an adversarial reviewer to name the most tempting alternative missing-state interpretation and why it is wrong. A claimed model confidence score is not itself documentary proof of semantic fidelity. For controlled codes, favor deterministic mapping tests and human review of the visible explanations.
When the source is defective
Sometimes a field is genuinely ambiguous: an old database uses N/A for both inapplicable and unavailable, or a paper table uses identical dashes for several reasons. A translator cannot recover facts that the source never encoded. Preserve the ambiguity transparently, propose a controlled disambiguation, escalate to the data owner and document the resolution. If publication cannot safely proceed, a neutral unresolved label or a withheld release is preferable to fabricated specificity. Translation fidelity includes knowing when translation alone cannot repair the source.
11. Frequently asked questions: missing-value translation and localization
FAQ 01. Is a missing value the same as zero?
No. Zero is a numerical value and often a meaningful observation or computed result. Missing means the source does not provide a value under the relevant representation rule. Some badly designed datasets use zero as a missingness sentinel; that convention must be interpreted using its own codebook, not treated as universal.
FAQ 02. Is null the same as a blank field?
Not necessarily. In JSON, null is a literal value. A blank interface cell or empty CSV text may be mapped to null in a particular system, but that behavior comes from schema and import rules. Translation should preserve the defined state and distinguish it from appearance alone.
FAQ 03. Is null the same as an omitted JSON property?
No. JSON Schema specifically distinguishes null from absence. Some applications assign different operational meanings to these two cases. A translator of API documentation should maintain the syntax and explain each case according to the documented application contract.
FAQ 04. What does N/A mean in translated documents?
It commonly suggests not applicable, but it can also be used informally for not available. Verify the source legend or data dictionary before expanding it. When the source mixes meanings, ask for clarification or keep a carefully scoped ambiguous status rather than inventing precision.
FAQ 05. What is the difference between unknown and not applicable?
Unknown usually refers to a value that might exist or be expected but is not known to the source. Not applicable means that there is no proper value for the field in the given case. The latter can legitimately cause a form question to be skipped; the former may require later follow-up.
FAQ 06. Does a blank survey answer mean no?
No. A blank may indicate that no choice was made, the question was not shown, or the answer failed to save. An explicit negative answer is a distinct event. Consent forms and decision surveys should preserve the unselected state in both storage and language.
FAQ 07. How should prefer not to say be translated?
Preserve the respondent’s deliberate decision not to disclose, with natural and respectful phrasing in the target language. Do not equate it to not knowing the answer or to an accidental omission. Follow the survey’s coded choices and any relevant privacy requirements.
FAQ 08. What does not asked mean?
The information was not solicited through the relevant workflow. It differs from asked but unknown, asked but declined and no response after a question was presented. In forms with branching logic, a not-asked code may be needed to interpret response rates and eligibility correctly.
FAQ 09. Can a system insert zero when a number is missing?
Only if a documented business or statistical rule explicitly authorizes that default or imputation, and the resulting value is identified appropriately. A translator should not implement the rule independently as a wording change. Silent zero substitution can alter aggregates, eligibility and decisions.
FAQ 10. Is no data available the same as no cases occurred?
No. No data available states a limitation of evidence or access. No cases occurred is a claim about events in the world. A completed surveillance process may support a recorded count of zero; an outage or absent report does not.
FAQ 11. What does pending mean in a translated status?
A process is unresolved or incomplete at a stated point in time. The translation should not turn it into approved, failed, negative or permanent absence. If an anticipated completion date is not given, do not invent one.
FAQ 12. What is the difference between withheld and unknown?
Withheld indicates deliberate nonrelease of information, whereas unknown identifies a knowledge gap. A holder might know exactly what is hidden. Translating withheld as unknown can conceal the disclosure reason, while translating unknown as withheld invents a deliberate act.
FAQ 13. What is the difference between masked and redacted?
Masked commonly refers to concealing parts of a value in a display or system presentation; redacted commonly describes content obscured or removed from a released copy. The exact institutional definitions should govern. Neither automatically means the source record never contained the information.
FAQ 14. How do I translate no record found?
State the result of the actual search, with its scope when needed. A search that finds no match under one identifier does not prove that no relevant record exists anywhere. Preserve whether the system completed the search or failed to retrieve results.
FAQ 15. What does below detection limit mean?
It means the measured signal or quantity falls below the stated method’s relevant detection threshold. It is not necessarily zero. The translation must retain the limiting relation, units and methodological qualifications, with specialist review where a technical standard governs reporting.
FAQ 16. How should a dash in a table be translated?
Check the legend and data dictionary before copying or expanding it. A dash can represent missing, not applicable, suppressed, zero or other editorial conventions depending on the publication. In accessible formats, provide meaningful text rather than relying on unexplained punctuation.
FAQ 17. Should I translate machine-readable absence codes?
Usually controlled tokens, API keys and schema-defined code values stay unchanged unless a formal migration or localized data layer specifically authorizes mapping. Translate human-facing descriptions and documentation instead. Validate that exported codes and downstream integrations still work.
FAQ 18. What is the difference between missing and unsupported?
Missing refers to absent information; unsupported describes a capability limitation in a system, field or version. A value may exist and apply to an entity while an old client cannot show it. A translation should preserve that technical constraint rather than declaring the property irrelevant.
FAQ 19. Can AI translation invent a missing value?
Yes, automated translation or summarization can produce a fluent but unsupported number, date or status. Compare all data-bearing claims with their source records, especially after paraphrasing. A claim that cannot be traced to source evidence should be queried or removed.
FAQ 20. How are missing values handled in charts?
Preserve the source state in both data and visual explanation. Do not draw a missing point at zero or connect across a gap as though a measurement existed unless the chart’s documented method permits interpolation. Check tooltips, legends, exports and accessible descriptions.
FAQ 21. What does imputed mean?
An imputed value is supplied through a stated analytical method in place of an unavailable observation. It is not itself an observed source fact. The translation should preserve the methodological marker and avoid hiding assumptions about uncertainty.
FAQ 22. How should not performed be translated in a medical report?
It must remain distinct from a completed negative or normal result. Use the relevant health system’s approved wording and codes, and obtain appropriate professional and linguistic review. A translation should never generate a clinical conclusion from an unperformed procedure.
FAQ 23. What if the source uses one word for several absence reasons?
Do not choose a more specific target meaning without evidence. Inspect the surrounding workflow and metadata; if the distinction remains unresolved, preserve ambiguity or raise a query. Fabricating a precise reason is often more damaging than admitting that the source is imprecise.
FAQ 24. What if the target language has no short equivalent?
Use a concise explanatory phrase, or an approved short label with accessible help text. Natural language does not require one-to-one word matching. The target must preserve the meaningful state and the correct reader decision, even when it takes more words.
FAQ 25. How should a translated form be tested?
Test the entire workflow with actual state combinations: unselected, explicit yes, explicit no, refused, unknown, not asked, system error and inapplicability where allowed. Inspect stored values, display labels, validation, exports and assistive-technology output. The goal is functional equivalence, not just good wording.
FAQ 26. When is a source query mandatory?
When the distinction changes eligibility, money, safety, legal responsibility, statistical conclusions, privacy or a user action and the source does not settle it. Keep the query auditable and do not replace unanswered questions with ungrounded assumptions.
FAQ 27. What distinguishes the present chapter from numerical fidelity?
The preceding numerical chapter covers units, precision, estimates, tolerances, percentages and number formatting when numerical values are present. This chapter covers whether a value is present, missing, inapplicable, unknown, restricted, pending or only displayed as blank. Their quality checks complement rather than duplicate each other.
FAQ 28. Can I calculate totals after translating a missing-value label?
Not simply because the label changed. Computation depends on the dataset’s documented missingness treatment and actual numeric inputs. Translators should preserve the source classification and ask analysts to verify whether missing, zero, suppressed and estimated values may enter an aggregate.
FAQ 29. Is not collected the same as not reported?
No. Not collected concerns whether the information was gathered. Not reported concerns what was included in a report. Data can be collected without being publicly reported. The translation must not make claims about upstream activities that the publication does not establish.
FAQ 30. What is the final release criterion?
A target-language reader, reviewer and downstream system should distinguish every state the source distinguishes, make the same permitted inferences and take the same appropriate actions. Test data type, terminology, grammar, user interface, export and disclosure behavior. Record unresolved ambiguities rather than concealing them.
12. Source notes and next reading routes
Technical data representation: The IETF RFC 8259, The JavaScript Object Notation Data Interchange Format defines JSON’s value types; JSON Schema’s null reference distinguishes null from an absent property. These references ground examples of machine representation, but they do not define the semantic meaning of each application’s null value.
Data absence in structured records: HL7 FHIR DataAbsentReason is a high-authority example of distinct reasons for absent values. W3C’s Model for Tabular Data and Metadata on the Web describes a tabular semantic model and its treatment of empty cells. Domain systems may implement different conventions; check each governing version.
Language and evaluation: Unicode CLDR’s numbers specification guides locale-sensitive number and plural rendering. MQM Core and the MQM Error Typology provide categories for translation-quality review. They support, but do not replace, source owner clarification, legal review, statistical analysis or expert validation.
Continue with Master Art of Translation | The Complete System for Moving Meaning Between Languages to revisit source analysis, equivalence and the complete translation architecture. Read the preceding Measurement and Uncertainty Fidelity System when the original value exists but its numerical expression, unit, precision or uncertainty must be retained.
For the English meanings and collocations of words such as absent, missing, pending, refuse, unknown and withhold, the protected Vocabulary Learning Hub is the appropriate independent reading route. For how English clauses, negation, passive constructions and status expressions work, consult How English Works | Translation and Meaning Across Languages. Neither of those owners is replaced by this translation-specific chapter.
Final principle: a blank-looking cell is not a fact about the world. Find the data state, its source and its boundaries; preserve it through translation, display, computation and review; and do not offer the reader more certainty than the source warrants. The same rule protects a simple customer form and a complex international data exchange.