Biodiversity Data & Darwin Core Translation

How do you translate biodiversity data, scientific species names, specimen labels or GBIF occurrence records without corrupting the evidence? The answer is not to replace every term with a natural-sounding equivalent. A dependable biodiversity translation workflow must preserve the identity of each taxon, the source observation, the time and location of a record, the certainty of an identification, and the exact relationship between a human-readable description and its machine-readable fields. This guide explains Darwin Core translation, taxonomic nomenclature, biological data localisation, species occurrence records, natural-history collections and multilingual conservation communication as one carefully controlled meaning-transfer system.

People searching for how to translate scientific names, what Darwin Core fields mean, how to localise GBIF datasets, how to translate specimen data into another language, or how to distinguish accepted names from synonyms need more than a dictionary. A Latin scientific name is not an ordinary sentence. A species occurrence is not automatically a living animal counted today. An observation, preserved specimen, material sample, checklist assertion and sampling event represent different kinds of evidence. The central problem is deciding which words should be translated, which identifiers must be preserved, and which scientific claims require qualification.

This world-facing Master Art of Translation chapter provides a repeatable method for translators, researchers, data managers, museum curators, ecology teams, software developers, conservation communicators and multilingual readers. It moves from search intent and hidden data problems to the mechanism, a production workflow, failure modes, worked fictional diagnostics, quality gates, transfer exercises and trustworthy next routes. Its technical anchors are TDWG Darwin Core, GBIF documentation, 2026 taxonomy-migration guidance and established translation-evaluation standards; it does not change any existing vocabulary or English-system article, and it does not present invented demonstration records as real field observations.

Architecture node: EDKSG-TRANS-MASTER-WORLD-630. Canonical parent: Master Art of Translation | The Complete System for Moving Meaning Between Languages. Scope: translation of biodiversity record meaning, not the ownership of biological taxonomy, general English instruction, or vocabulary acquisition.

The answer in one minute: translate meaning, protect evidence

Read the source as a structured claim, not an isolated sentence. Determine whether its object is a taxon, an organism, an occurrence, a field event, a specimen, a sample, an identification, a spatial reference or a publication. Identify the canonical field names and identifiers, keep those machine-facing elements stable, and translate only the display labels, explanations and approved free text where the project permits it. Verify facts with the scientific source, then run an independent field-by-field comparison, especially for negation, uncertainty, time, quantity, locality, taxonomic status and permissions.

A simple thought experiment shows why. A record may say a taxon was reported as present during a survey, but it may not mean that a current population exists, that an organism was captured, or that the identification was later confirmed. A translator who turns that record into an attractive natural-language claim such as ‘the species thrives here’ has added abundance, time and ecological interpretation that are absent from the evidence. The target text sounds polished while the scientific meaning is wrong. The correct transfer retains the modest statement and flags what remains unknown.

Think of the work as five separate layers: a persistent record identity; an observation or sampling event; a taxonomic interpretation; an environmental and geographic context; and a target-language explanation. Each layer carries different kinds of certainty. The most useful general test is whether an independent reader could reconstruct the original dataset’s claims from the translation without acquiring extra facts, losing qualifications, or confusing database keys with ordinary vocabulary.

Why a specialist translation architecture is necessary

Ordinary translation can often preserve a message by reformulating sentences in natural target grammar. Biodiversity records impose a second constraint: they are also computational objects. Software may join rows through identifiers, count occurrences, display maps, compare checklists, and infer changes through time. Altering a field value because it looks like English may break those operations. Conversely, leaving every human-facing note untranslated can make the dataset inaccessible to its intended readers. The art is controlled separation, not blanket translation or blanket preservation.

A specimen record can contain a handwritten historical label, a modern transcription, a curatorial correction, a taxonomic name accepted under a selected checklist, a locality with a controlled disclosure policy, and a date inferred from imperfect writing. These are not interchangeable versions of the same claim. They are distinct statements with distinct provenance. If the target text silently replaces the historical label with the newest hypothesis, it changes the archival evidence. If it translates a locality too freely, readers may lose the original place reference. If it removes a question mark, uncertainty disappears.

The biological reference system is also evolving. TDWG’s Darwin Core list of terms identifies a version issued on 13 October 2026. GBIF announced on 23 June 2026 that its website had adopted the Catalogue of Life Extended Release as its default taxonomy, while some API behaviours and identifiers required explicit checklist context. This makes it especially unsafe to assume that an older numerical taxon key and a newer alphanumeric key are interchangeable labels. A good 2026-era guide teaches readers to identify the classification authority and version before comparing outputs.

There is an ethical dimension too. Some observations describe threatened taxa, nesting sites, community-shared knowledge, genetic resources, or privately owned collections. Translating hidden coordinates into a more precise public address, dropping a disclosure restriction, or representing culturally specific names as universal equivalents can cause harm. Language work is therefore part of data governance. The translator is not authorised to resolve scientific, privacy or conservation policy questions simply because the target sentence would read more smoothly without them.

The eight invariants a biodiversity translator must protect

Record identity: persistent identifiers and joins must point to the same objects after translation.
Taxon interpretation: names, accepted usages, synonyms and taxon concepts must remain distinguishable.
Observation status: recorded presence, true recorded absence, non-detection and no available data are different.
Evidence lineage: specimen, human observation, machine observation, material sample and inferred statement retain their sources.
Space and time: coordinates, uncertainty, datum, dates, time zones and historical precision cannot drift.
Quantities and methods: count, effort, units, protocols and detection limits remain explicit.
Qualifications: tentative identification, revisions, scope restrictions and unresolved disagreements survive.
Rights and safety: licence terms, attribution, sensitive-locality generalisations and community restrictions follow the data.

The standards and authorities behind the method

Darwin Core is the controlled biodiversity vocabulary maintained through Biodiversity Information Standards (TDWG). Its purpose is to facilitate sharing information about biological diversity, including taxa, occurrences, events, identifications, specimens, measurements and references. It is not a rule that every field value must be translated. Its term identifiers and semantics provide a shared cross-language anchor; readable labels and prose can be localised while structural fields stay stable. Source: TDWG Darwin Core standard and the Darwin Core term list, issued 13 October 2026.

GBIF’s Integrated Publishing Toolkit guidance distinguishes resource metadata, checklist data, occurrence data and sampling-event data. In practical translation, that means the first decision is which kind of dataset is present and how its records relate. GBIF also explains taxonomic matching, names, synonyms and flags: an index match is an interpretation used for discovery, not proof that a specimen’s original identification must be rewritten. Source: GBIF IPT publication manual, GBIF taxonomy interpretation and GBIF Species Matching tool, published April 2026.

The 23 June 2026 GBIF migration notice explains the shift from the legacy GBIF Backbone to Catalogue of Life Extended Release. For translation and documentation, it is essential to record which checklist a taxonKey belongs to and whether an identifier is numeric or alphanumeric; do not silently migrate fields inside an evidence-preservation job. Source: GBIF taxonomy migration guidance, 23 June 2026. Humboldt Extension work, ratified by TDWG on 28 February 2024 and incorporated into Darwin Core, provides additional vocabulary for inventory effort and completeness; source: TDWG Humboldt Extension.

A translation review can use the ideas of bilingual comparison, purpose fitness and classified errors from ISO 5060:2024, which covers evaluation of human, post-edited and unedited machine translation output. The service workflow perspective is informed by ISO 17100:2015. This chapter does not claim ISO certification or reproduce proprietary standard text. It adapts the basic principle that technical translation errors have different consequences and should be evaluated against a documented brief.

The repeatable production workflow: twenty decisions in order

01. Confirm the evidence object

Classify each item as a Taxon, Occurrence, Organism, Event, MaterialSample, Identification, Location, Measurement, collection record or explanatory article. The noun determines which claims are permitted.

If the class is unknown, request the data dictionary and preserve an explicit unknown rather than guessing from nearby words.

The operational hand-off is a small evidence note: record the source field or document section, the rationale for the language decision, the reviewer responsible for any specialist judgement, and the exact check by which a future curator could reverse or reproduce the decision. This converts a one-off translation into a maintained publishing process rather than an untraceable collection of fluent sentences.

02. Define the receiving reader

A researcher downloading a CSV, a curator searching an accession, a citizen reading a map and a student reading a species profile need different display layers.

The translation brief should name its audience and distinguish searchable machine values from readable labels before any edits.

03. Freeze the source version

Record dataset DOI or persistent URL, publisher, export timestamp, schema version, change log and the exact file checksum if supplied.

A translation cannot be reliably audited when the upstream rows change silently during the language project.

04. Select an authority for names

Identify the scientific checklist or taxonomic treatment, its date, identifiers and handling of split or lumped taxa.

Do not substitute a new accepted name on the strength of casual browsing; add a separate mapped interpretation if authorised.

05. Extract the identifier inventory

List occurrenceID, taxonID, organismID, eventID, materialSampleID, catalogNumber and all relationship fields.

Treat those as identity-bearing strings; preserve case and punctuation unless the data owner’s schema explicitly authorises normalization.

06. Inspect file structure

Determine whether the asset is a spreadsheet, database view, Darwin Core Archive, EML metadata file, public caption or narrative report.

Protect column names, encoding, delimiters, links and schema references when translating the prose surrounding them.

07. Map field semantics

Build a table linking each original header to its TDWG definition, approved human-readable label, data type, allowed values and locale notes.

A localised display label is not permission to rename the canonical key in an interoperable export.

08. Mark controlled vocabularies

Separate enumerated values such as basisOfRecord, occurrenceStatus and taxonomicStatus from unconstrained explanatory text.

Keep machine-readable codes in the data layer and display a verified translated explanation outside the code where useful.

09. Audit nomenclature

Check scientificName, authorship, rank, qualifier, name usage, synonym relationship and taxon concept against the chosen authority.

Italic style is presentation, not evidence that a name is valid; do not infer validity from typography.

10. Identify uncertain evidence

Locate cf., aff., question marks, identificationVerificationStatus, date estimates, georeference notes and inherited catalogue assumptions.

Target-language fluency must never remove the distinction between what a person saw and what later reviewers inferred.

11. Protect coordinates and units

Check latitude, longitude, coordinateUncertaintyInMeters, geodeticDatum, elevations, depths, measurementUnits and numerical precision.

Convert presentation formats only under a documented mapping that preserves values and lets users reconstruct the original.

12. Distinguish temporal layers

Separate eventDate, verbatimEventDate, identificationDate, publication date, download date and the date a record was modified.

The date of the database export is not the date on which a specimen was collected or a species was observed.

13. Preserve survey design

Read samplingProtocol, samplingEffort, eventID hierarchy, sample size and inventory completeness fields before interpreting counts.

Do not turn an unsearched habitat into an absence record or a single observation into a population estimate.

14. Review rights and sensitivity

Inspect licence, owner, attribution, informationWithheld, dataGeneralizations and applicable access arrangements.

A translated public description must not disclose a protected locality that the structured dataset intentionally obscures.

15. Draft the human layer

Translate labels, accessible descriptions, help text, explanatory notes and approved narrative fields using natural target-language grammar.

Place limitations close to the claim they qualify; do not bury important uncertainty in a distant glossary.

16. Run terminology alignment

Check repeated biological concepts across tables, maps, specimen cards, API documentation and narrative pages.

A species profile should not alternate between occurrence and population, or sample and specimen, because translators worked on different assets.

17. Perform structural QA

Compare row counts, IDs, nulls, headers, numeric fields, relationship keys, language tags and escaping after localisation.

A file that reads well but no longer joins to its source tables has failed the brief even if the prose is attractive.

18. Perform scientific QA

Have a qualified subject reviewer inspect name interpretation, evidence class, uncertainty, geography, sampling logic and restrictions.

A bilingual editor can catch language errors but may not know whether a biological inference is scientifically supportable.

19. Evaluate target usability

Ask readers to find a record, understand an uncertain identification, distinguish an observation from a specimen and trace the source.

If the interface makes the user conclude more than the original record warrants, correct the explanation.

20. Version and publish carefully

Deliver the target content with source attribution, translation date, schema version, authority choices, unresolved questions and limitations.

Publish the translated explanatory layer without changing the canonical data or promising that later upstream taxonomic updates are already reflected.

Forty-eight mechanisms: where biodiversity meaning really lives

Occurrence versus organism

Mechanism. An occurrence records evidence of a taxon at a place and time, or equivalent event context; an organism can persist through multiple occurrences.

Hidden translation failure. Calling one occurrence an individual animal makes repeated sightings look like multiple animals. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Translate occurrence as a record of presence or evidence when that is what the data actually contains. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Reconcile occurrenceID against organismID and inspect whether one organism appears repeatedly. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Taxon versus scientific name

Mechanism. A taxon is a circumscribed biological concept under a particular treatment; a scientific name is a linguistic and nomenclatural label.

Hidden translation failure. A neat translated common name may refer to several taxon concepts. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep name strings separate from taxon identifiers and selected taxonomic authority. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check taxonConceptID, scientificNameID and the checklist version before proclaiming equivalence. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Accepted name versus synonym

Mechanism. A checklist may mark one name usage accepted and another a synonym, without erasing historical labels.

Hidden translation failure. Translating both records as the accepted name destroys the link to prior literature and specimen labels. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve originalNameUsageID or acceptedNameUsageID relationships and explain synonyms as relationships. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Test whether a reader can still find records under the old usage and the current interpretation. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Observation versus collection

Mechanism. HumanObservation is a record of observation, whereas PreservedSpecimen describes a different evidence path.

Hidden translation failure. Replacing both with ‘sample’ obscures whether a physical object was collected and curated. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Use the controlled basisOfRecord and translate supporting text accordingly. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Ask what physical voucher exists and whether its identifier is cited. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Material sample versus organism

Mechanism. A tissue tube, soil subsample or eDNA extract is not equivalent to a whole living organism.

Hidden translation failure. A laboratory extract with one barcode can be mistaken for one animal in an abundance report. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve materialSampleID and sampling relationship; explain sample type, not imagined individuals. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check whether organismQuantity is measured, inferred or absent. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Identification versus observation

Mechanism. An observer may record a bird while another specialist assigns a later species determination.

Hidden translation failure. Collapsing observation and identification dates can make a revision appear contemporaneous with the event. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep identifiedBy and dateIdentified distinct from recordedBy and eventDate. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Trace who asserted each taxon name and when. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Evidence versus inference

Mechanism. Photographs, audio clips, voucher specimens and unverified notes support different levels of conclusion.

Hidden translation failure. A caption stating confirmed presence may overstate what an ambiguous sound recording proves. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Translate the strength and provenance of the evidence without inventing verification. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Require independent support for any move from probable to confirmed. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Occurrence status versus missing data

Mechanism. An explicit absent value asserts documented absence under some observation regime, not a simple missing row.

Hidden translation failure. Turning no record into absent can falsify a species distribution map. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Retain status values and explain survey conditions before treating non-detection as absence. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Compare effort and detection methods when interpreting negative findings. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Count versus abundance

Mechanism. One record of several individuals is not the same as several records of one individual.

Hidden translation failure. A translated chart that calls number of rows population size induces serious ecological error. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve individualCount or organismQuantityType and explain what was counted. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Try deduplicating by organism and event; check whether the claimed number survives. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Life stage versus age

Mechanism. Larva, juvenile, adult and reproductive stage can follow controlled vocabulary rather than calendar years.

Hidden translation failure. Calling all juveniles ‘young adults’ changes biological classification and survey comparability. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Use source discipline terms and maintain the controlled lifeStage value. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check the organism group because stage labels differ across taxa. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Sex versus gender label

Mechanism. A biological sex field should not be treated as an invitation to impose human social categories or unrecorded assumptions.

Hidden translation failure. Editorial smoothing can change unknown into a positive classification. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Follow the record’s biological definition and preserve uncertain or inapplicable values. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Ask whether the source assertion was observed, inferred or not collected. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Habitat versus locality

Mechanism. Habitat describes environmental context; locality names a place, sometimes with ambiguous historical labels.

Hidden translation failure. Translating ‘reedbed’ as a named wetland creates false geographic precision. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep descriptive habitat prose apart from location identifiers and coordinates. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Confirm whether a phrase is a vegetation description or a proper place name. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Coordinates versus address

Mechanism. Latitude and longitude express geospatial data, not a postal address or exact point of collection in every record.

Hidden translation failure. A centroid rounded to one decimal place may be interpreted as a precise nest location. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Protect coordinate system, uncertainty and data generalization details. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Map the uncertainty radius before writing point-specific prose. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Geodetic datum versus map style

Mechanism. The reference frame gives coordinates their geospatial meaning; it is not a cosmetic map setting.

Hidden translation failure. A translated field label can imply all points use WGS84 when older records have other datums. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Retain geodeticDatum and document any actual reprojection outside translation. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check whether values and units were transformed or only their labels changed. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Coordinate uncertainty versus precision

Mechanism. Decimal places do not automatically measure positional certainty; reported uncertainty has its own field and method.

Hidden translation failure. Rounding a coordinate to more digits visually creates confidence unsupported by the source. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Explain coordinateUncertaintyInMeters as a documented spatial uncertainty measure. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Compare source notes and georeferencing protocol, not the number of printed decimals. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Date recorded versus date published

Mechanism. A collector’s field date, a determination date and a repository release date describe different events.

Hidden translation failure. Writing ‘recorded in 2026’ from a 2026 export may recast a century-old specimen as recent. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep eventDate, dateIdentified and publication timestamps distinct. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Cross-check the history of the underlying occurrence with catalogue changes. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Verbatim date versus interpreted date

Mechanism. A damaged label reading ‘early spring 19–‘ is historical evidence that may be normalized only with stated assumptions.

Hidden translation failure. Replacing uncertainty with a calendar day invents observation precision. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve verbatimEventDate alongside interpreted eventDate and record derivation details. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. A reviewer should reconstruct the original inscription without guessing. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Sampling event versus specimen row

Mechanism. An event connects time, place, sampling protocol and potential multiple taxa or occurrences.

Hidden translation failure. Treating each row as a unique field visit inflates survey effort. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve eventID relationships and explain nested events when applicable. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Count unique events separately from occurrence records. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Sampling protocol versus technique label

Mechanism. A net type, trap duration, transect method or acoustic detector setting conditions what the data can show.

Hidden translation failure. A broad translation such as ‘wildlife survey’ may conceal meaningful detection bias. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep method specifics and published protocol identifiers where available. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Ask whether another team could reproduce the observation conditions. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Sampling effort versus number of finds

Mechanism. Effort refers to how much searching was done, not how many organisms were detected.

Hidden translation failure. A translation that equates ’20 trap-nights’ with ’20 insects’ changes the denominator of analysis. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep units such as hours, nights, quadrat area or line length attached to effort. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Test any reported encounter rate with the original denominator. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Completeness versus total abundance

Mechanism. An inventory completeness statement describes coverage of a target group under an explicit sampling design.

Hidden translation failure. Writing ‘all species were found’ is unjustified when completeness is partial or unknown. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve Humboldt-style completeness context and defined survey universe. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check targeted taxon set, effort and completeness definitions. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Taxonomic rank versus literal word

Mechanism. Species, subspecies, variety and form have nomenclatural consequences and should not be translated as informal descriptors.

Hidden translation failure. The word variety in a botanical scientific name cannot be casually reduced to ‘kind’. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Protect scientific rank markers and use approved display equivalents. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Verify rank abbreviations against the relevant code and source name string. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Scientific authorship versus contributor

Mechanism. Name authorship records nomenclatural authority, not necessarily the person who identified a specimen.

Hidden translation failure. Translating an author abbreviation as a collector changes attribution. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve scientificNameAuthorship and link it to nomenclature, not observation provenance. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Compare author, recordedBy and identifiedBy separately. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Name qualifier versus certainty claim

Mechanism. cf., aff. and other qualifying marks express identification relationships or doubt; conventions vary by discipline.

Hidden translation failure. Deleting cf. makes an uncertain identification look resolved. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep qualifier notation or translate its function explicitly when a user-facing explanation is required. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Inspect identificationQualifier and the actual determination note. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Vernacular name versus scientific name

Mechanism. One species can have many community names and a shared everyday name can refer to several species.

Hidden translation failure. Picking an attractive common name as a one-to-one synonym can mislead target readers. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Record language, region, community use and taxonomic scope for vernacular labels. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check whether the proposed name is attested for the selected taxon concept. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Local knowledge versus universal catalogue

Mechanism. An ecological name may encode harvesting practice, season, relation or cultural responsibility.

Hidden translation failure. Flattening a local term into a generic English label can remove knowledge, consent and context. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Retain provenance and consult appropriate knowledge holders before public adaptation. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Audit consent, attribution and whether the label was intended for broad redistribution. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Species concept versus checklist mapping

Mechanism. Different classification systems may lump or split organisms under different taxon boundaries.

Hidden translation failure. A translation that adopts one modern name without context can silently alter study comparisons. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. State ‘according to’ the named authority and its date where classification matters. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Compare the source concept with the target authority’s circumscription. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Endemic versus native

Mechanism. Endemic describes a restricted distribution; native means occurring within an ecological range without introduction under a given definition.

Hidden translation failure. The two are not synonyms and neither establishes conservation status. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep each claim tied to evidence, range and jurisdictional or biogeographic scope where relevant. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check the data source for the exact geographic basis without inventing boundaries. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Introduced versus invasive

Mechanism. Introduced records a pathway or non-native establishment; invasive describes a further impact or policy classification.

Hidden translation failure. Calling every introduced plant invasive can exaggerate environmental risk. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep establishmentMeans, degreeOfEstablishment and pathway assertions distinct. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Confirm the criterion and authority used before translating an impact label. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Threatened versus rare

Mechanism. Population rarity and formal extinction-risk categories measure different things.

Hidden translation failure. A taxon listed as uncommon cannot be promoted into a formally threatened category through a headline. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve referenced assessment category and date, not an intuitive adjective. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Identify whether the label comes from a formal list or informal observation. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Historical specimen versus current range

Mechanism. A century-old collected specimen is evidence for a past occurrence and not necessarily present occupancy.

Hidden translation failure. A translated species map titled ‘where it lives today’ can mislead when it includes historic records. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Keep temporal filters explicit and describe the map as records rather than current population. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check time intervals and whether the most recent evidence is independently confirmed. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Literature assertion versus new observation

Mechanism. A checklist may cite a species reported in a paper without supplying a new field occurrence.

Hidden translation failure. Rewording the citation as a recent observation creates a new false evidence event. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Maintain citation context and record the source’s use of literature. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Trace the reference back to the original study or specimen. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Unpublished identification versus formal description

Mechanism. A provisional label, working name and officially published nomenclatural act have different standing.

Hidden translation failure. Putting a provisional name in italics does not make it validly published. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Explain provisional status and avoid claims about code-compliant publication without evidence. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check publication metadata and formal nomenclatural criteria. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Image metadata versus sighting proof

Mechanism. A photograph may show a taxon but camera time, place and geotag can be edited or absent.

Hidden translation failure. Treating metadata as independently verified evidence overstates what the file alone supports. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Translate descriptive captions while keeping media provenance and limitations visible. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check whether photo ID, location and verification trail are linked. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Audio event versus animal count

Mechanism. A recording may capture calls from uncertain numbers of individuals.

Hidden translation failure. Translating 12 detections as 12 animals may inflate abundance. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Maintain detection and classification units and distinguish event frequency from individual identity. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Review detector thresholds and annotation quality before reporting totals. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

eDNA signal versus live presence

Mechanism. Environmental DNA can indicate genetic material without directly counting living organisms at an exact coordinate.

Hidden translation failure. Calling one sequence detection an observed animal is a categorical error. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Explain laboratory method, contamination controls, detection threshold and uncertainty. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check whether a voucher or visual observation exists independently. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

DNA barcode versus species certainty

Mechanism. A barcode identification depends on reference sequences, marker coverage and matching thresholds.

Hidden translation failure. A high similarity score is not universal proof of a unique species determination. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve the algorithm’s match and confidence language, not an invented certainty. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Review the reference database and marker limits. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Reference dataset versus target dataset

Mechanism. A taxonomy matching service supplies an interpretation relative to its own reference checklist.

Hidden translation failure. Translation can accidentally overwrite verbatim source names with index-side interpretations. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Store match result separately from original taxon assertion. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Test whether the original source strings remain recoverable. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Historic place names versus contemporary geocoding

Mechanism. A specimen label may use a locality name that has changed, broadened or become ambiguous.

Hidden translation failure. Replacing the verbatim place name with a present-day guess erases archival detail. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve verbatimLocality and add a qualified interpreted locality separately. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Ask what documentary support underlies a modern-coordinate assignment. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Uncertainty note versus empty field

Mechanism. A missing uncertainty value means unknown or unavailable, not necessarily zero uncertainty.

Hidden translation failure. Leaving out a caveat while using a precise-looking map point can create false trust. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Display uncertainty where present and explain what unknown does not mean. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Check null handling against map defaults and client-side fallbacks. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Data provenance versus editor improvement

Mechanism. A correction may be sound but still needs attribution and a traceable change record.

Hidden translation failure. Silent correction in translation makes datasets diverge and obscures who changed an assertion. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Flag the correction for the source owner and keep its status separate from translated text. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Compare modification history, contributor and issue record. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Public accessibility versus expert compactness

Mechanism. A technical label may be exact but unreadable to a broad audience.

Hidden translation failure. Over-simplifying can erase the qualifiers that justify interpretation. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Provide a plain-language explanation alongside the canonical field and preserved uncertainty. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Ask a reader to restate what the record supports and does not support. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

License versus access permission

Mechanism. A record’s reuse licence, embargo, attribution and publication restrictions are not equivalent.

Hidden translation failure. A ‘downloadable’ item is not automatically available for unrestricted reuse. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Translate rights notices precisely and retain licence identifiers and attribution obligations. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Test a proposed reuse against the actual source licence and policy. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Null, zero and ‘unknown’

Mechanism. No value, a measured zero and an explicitly unknown category carry different claims.

Hidden translation failure. Replacing blanks with zero can fabricate absence or measurement results. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Preserve null states in data; explain missingness only where permitted. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Run type-checked comparisons for every field after export. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Human translation versus AI output

Mechanism. Generative output can produce plausible names, dates and localities not present in a record.

Hidden translation failure. A polished target paragraph can hide hallucinated specimen or conservation claims. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Use machine assistance only under an evidence-locked brief with verification. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Compare every scientific assertion, identifier and number to the source. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Data service versus public narrative

Mechanism. An API payload and a public-facing species page have different translation targets.

Hidden translation failure. Changing API field keys for human readability can break consumers while leaving the interface looking correct. The resulting sentence may be easy to read yet hard to reconcile with the dataset, because a surface word has acquired a scientific role that the original evidence did not give it.

Repeatable decision. Localise display strings and document schema separately from canonical machine endpoints. Where a brief requires explanatory adaptation, put the explanation in a documented human-facing layer and keep the source assertion and machine fields independently recoverable.

Diagnostic. Validate both API conformance and reader comprehension in the final application. A reviewer should also try a reverse reading: what exact source assertion would a reader reconstruct from the translated sentence? A stronger claim, different record type, broader spatial scope or greater certainty is a substantive error, not a stylistic choice.

Worked translation diagnostics: eighty fictional field situations

The following are deliberately fictional source snippets and training cases, not claims about actual observations, locations, collections or protected species. The useful exercise is to identify the one part of meaning most likely to drift and write a target-language explanation that a scientist could audit. In real projects, source owners and authoritative checklists decide contested scientific facts.

A01 — one row, two sightings

Fictional source situation: The database has two occurrenceIDs but one organismID for a tagged bird photographed on different days.

Hidden failure: A summary calls the dataset ‘two birds’ because two rows are visible. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Translate as two records of the same tracked organism, with dates preserved; do not infer a population count. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Deduplicate by organismID and show both events; confirm that the public count changes from records to individuals. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A02 — specimen without living population

Fictional source situation: A preserved insect specimen was collected decades ago, then digitised this year.

Hidden failure: The target caption says ‘the insect currently lives in this area’. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Say the specimen documents a historical collection event, and state current presence only if separately supported. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare collection date with digitisation date and the evidence for continuing occupancy. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A03 — label transcription and correction

Fictional source situation: An old herbarium sheet has a faint locality transcription and a later curator’s clarified interpretation.

Hidden failure: One seamless target sentence hides which words were written on the sheet. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep the verbatim inscription and the modern interpretation in separate labelled fields. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Reconstruct the original text and attribute the later editor’s conclusion. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A04 — field note marked uncertain

Fictional source situation: An observer wrote ‘possibly species X’ beside a blurred photograph.

Hidden failure: An attractive translation removes ‘possibly’ and displays the taxon as confirmed. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Use ‘tentatively identified’ or another approved uncertainty label while retaining identificationQualifier. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Ask whether the original evidence supports definitive identification. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A05 — duplicate media

Fictional source situation: Three photographs show one whale surfacing during a single survey event.

Hidden failure: The captions are transformed into three sightings and then three whales. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep images as related evidence and document the single event rather than multiplying organisms. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Cross-check eventID and media links against occurrence counts. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A06 — catalogue number collision

Fictional source situation: Two institutions both use local catalogNumber 00421 for unrelated collections.

Hidden failure: A translator treats the local number as a global specimen identifier. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve institutionCode and collectionCode, and prefer persistent global identifiers when joining records. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Run a uniqueness test using the complete identifier context. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A07 — accession versus occurrence

Fictional source situation: A specimen has a museum accession reference and a separate occurrenceID shared with a data portal.

Hidden failure: The target label merges the identifiers and prevents external record lookup. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Give each identifier its proper role and use a human-readable label that distinguishes curation from observation. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Resolve both source links independently after publication. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A08 — voucher not available

Fictional source situation: A field observation was recorded without a physical specimen.

Hidden failure: The translation uses ‘collected sample’ merely because it sounds more scientific. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Describe an observation without implying capture or preservation. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Inspect basisOfRecord and evidence references before choosing sample-related nouns. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A09 — conflicting determination

Fictional source situation: Two specialists assigned different species names to a single preserved specimen in different years.

Hidden failure: The public description silently keeps only the newest name as if the earlier label never existed. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve the determination history, explain the selected current treatment and attribute each assertion. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check both dateIdentified and identifiedBy in the record history. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

A10 — reidentified archive item

Fictional source situation: An older specimen was originally described under a synonym now mapped to an accepted name.

Hidden failure: The translator overwrites the historical inscription with the accepted string. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain the verbatim scientificName and provide the new accepted usage as a separately sourced mapping. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Search both usages and verify that the original record remains discoverable. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B01 — Latin genus intact

Fictional source situation: A plant genus appears inside a multilingual explanatory paragraph.

Hidden failure: The translator changes the Latin name into a similar everyday noun. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve the scientific name’s spelling and typography, then explain any common name separately. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare character-by-character against the source nomenclatural string. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B02 — author abbreviation misunderstood

Fictional source situation: A binomial includes an abbreviated authority following the species epithet.

Hidden failure: The target describes that authority as the person who collected the specimen. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep nomenclatural authorship distinct from recordedBy and identifiedBy. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Inspect each field separately before writing contributor credits. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B03 — qualifier cf.

Fictional source situation: A determination reads Genus cf. species, indicating a comparative identification rather than settled equality.

Hidden failure: The target headline prints the species with no qualifier. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Carry the qualifier or an explicit cautious explanation and do not strengthen certainty. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check the record’s identification note and images before revising the label. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B04 — qualifier aff.

Fictional source situation: A biologist records an organism as resembling a named species without treating it as identical.

Hidden failure: The translator drops aff. and indexes the organism under the exact species. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Explain affinity or resemblance as the source intends, without converting it into definitive membership. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Ask the taxonomic reviewer whether the proposed mapping is supportable. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B05 — variety is rank

Fictional source situation: A botanical name includes var. to indicate a ranked infraspecific unit.

Hidden failure: The translated prose treats var. as an informal variety of colour. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Protect the rank marker and the full scientific name; explain rank only outside the nomenclatural string. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check the authoritative taxon record and the relevant naming convention. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B06 — cultivar is not wild taxon

Fictional source situation: A garden catalogue includes a cultivar epithet alongside a species name.

Hidden failure: The translator writes a nature report about a distinct wild species instead of a cultivated selection. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Separate cultivar nomenclature and trade name context from wild taxon concepts. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Review source intent and cultivar formatting before publishing. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B07 — homonymous genus

Fictional source situation: A short name matches multiple biological taxa in different kingdoms.

Hidden failure: A matching system chooses the wrong organism because the source’s kingdom was omitted in translation. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain higher classification and source authority when disambiguating names. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Test the same candidate name with and without kingdom and family hints. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B08 — split species

Fictional source situation: Two checklists differ on whether a wide-ranging taxon comprises one or several species.

Hidden failure: The target adopts a split without acknowledging that the source used the broader concept. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: State the taxonomic authority and preserve the source circumscription in comparative claims. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check taxonConceptID or treatment notes before joining distribution totals. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B09 — lumped species

Fictional source situation: A newer checklist combines two historical species under a broader accepted usage.

Hidden failure: Translation erases distinctions used in the original ecological study. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep the study’s original units and document the mapping as a later interpretation. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check that any pooled analysis is authorised by the research protocol. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

B10 — multilingual common name

Fictional source situation: The same everyday animal name is shared by several related taxa across languages.

Hidden failure: A translation supplies one local name as though it were the unique accepted species label. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Use scientificName as the stable cross-language key and qualify the vernacular name’s locale. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Have a native-language domain reviewer test whether the name is ambiguous. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C01 — absent is not unobserved

Fictional source situation: A survey checklist contains no row for a target orchid.

Hidden failure: The target narrative announces ‘orchid absent’ without any documented negative survey. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Write ‘not recorded in this dataset’ unless a true sampled absence is explicitly present. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Inspect sampling design, target group and negative detection records. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C02 — explicit absence

Fictional source situation: A fixed-point survey records occurrenceStatus absent after a documented protocol was applied.

Hidden failure: The translation softens the result into ‘no information available’. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve the recorded absence and specify the method and time boundary. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check protocol validity, sampling effort and uncertainty before generalising. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C03 — non-detection

Fictional source situation: An acoustic detector produced no target-species calls during a limited overnight period.

Hidden failure: The public map labels the habitat unoccupied. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Describe a non-detection under stated conditions, not proof that the species never occurs. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Review detector performance, season and weather as interpretation constraints. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C04 — present with uncertainty

Fictional source situation: A citizen observation was accepted only to genus level with an uncertain species suggestion.

Hidden failure: The translator displays a species-level icon as confirmed presence. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain the confirmed taxonomic rank and present the more detailed guess as tentative. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare taxonRank with identificationQualifier and verification trail. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C05 — several detections

Fictional source situation: A camera trap recorded six detection events of an unmarked mammal.

Hidden failure: The target report claims six individuals live there. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Translate as six detection events and avoid deriving independent individuals. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Inspect whether images, timing or identification distinguish the animals. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C06 — count interval

Fictional source situation: A field note gives ‘roughly 10–15’ flowering stalks in a quadrat.

Hidden failure: A concise table rounds the range to exactly 15 plants. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve approximation and distinguish stalks from plant individuals. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check counted object and interval representation in the visual table. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C07 — density per area

Fictional source situation: An ecological estimate reports nests per hectare rather than absolute nest numbers.

Hidden failure: A target infographic drops the denominator and multiplies apparent abundance. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep per-area units adjacent to the number and state the sampling area. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Recalculate the unit relationship using example values. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C08 — biomass versus count

Fictional source situation: A survey records dry biomass in grams rather than the number of organisms.

Hidden failure: The translation says ’30 individuals were collected’ from 30 g. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Name the measured quantity and leave counts unknown unless another field supplies them. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check measurementType, measurementValue and measurementUnit as a bundle. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C09 — observation by proxy

Fictional source situation: A track, scat or nest was recorded as evidence of a taxon without direct sighting.

Hidden failure: The target prose says ‘the animal was seen’. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Use ‘evidence consistent with the taxon was recorded’ and preserve evidence type. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check associated media and identification constraints. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

C10 — organism quantity type

Fictional source situation: A record says organismQuantity 42 with organismQuantityType percent cover.

Hidden failure: A generic translation claims 42 organisms in the quadrat. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain percent cover and explain how cover differs from individual count. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Test whether the numerical value can be interpreted without its quantity type. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D01 — latitude sign inversion

Fictional source situation: A CSV contains negative decimalLatitude values and positive decimalLongitude values.

Hidden failure: During localisation, the minus sign is removed as cosmetic punctuation. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep signed coordinates unchanged and separate display direction words from values. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Plot a sample point before and after conversion to detect hemisphere reversal. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D02 — longitude swap

Fictional source situation: A receiving form expects latitude before longitude, but the source has the reverse order.

Hidden failure: A user-facing translator copies numbers into the wrong slots while preserving both values. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Align field names, data types and coordinates rather than relying on visual column order. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Geospatially validate several records against the source uncertainty. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D03 — decimal comma

Fictional source situation: A locale uses commas as decimal separators but the downloadable CSV is comma-delimited.

Hidden failure: The target export changes 12.75 into 12,75 without properly quoting fields. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Localise the display, not machine numeric encoding; if exporting local decimals, specify delimiter and parser. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Import the file afresh and compare numerical values and row lengths. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D04 — uncertainty radius

Fictional source situation: A georeferenced collecting place has coordinateUncertaintyInMeters 5000.

Hidden failure: The target map hides the radius and describes an exact collection point. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Show the coordinate as approximate and preserve the declared uncertainty measure. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Zoom the map to the spatial error scale rather than its displayed decimal precision. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D05 — generalised rare-taxon point

Fictional source situation: A protected species record uses a deliberately coarse coordinate.

Hidden failure: The translated description reconstructs the hidden nest site from curator notes. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Respect informationWithheld and dataGeneralizations, and suppress sensitive detail in public text. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Have the conservation data steward approve disclosure independently of language quality. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D06 — altitude or depth

Fictional source situation: A sample is described as 200 m below sea level, but the target label reads 200 m elevation.

Hidden failure: Translation changes the vertical direction and environmental context. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep minimumDepthInMeters and maximumDepthInMeters separate from elevation variables. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check positive direction and reference frame in source metadata. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D07 — historical locality

Fictional source situation: An old label contains a place name with uncertain current correspondence.

Hidden failure: A translator replaces it with a modern settlement name without documentary support. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain verbatimLocality and identify any modern interpretation as provisional. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare archival gazetteers and collector itinerary notes when authorised. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D08 — place description and habitat

Fictional source situation: A record says ‘north-facing scree slope’ as habitat detail.

Hidden failure: The target stores ‘north-facing’ in a region-name field and loses the habitat feature. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep environmental descriptions in habitat or event remarks rather than geocoded administrative fields. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Ask whether the phrase names a place or describes terrain. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D09 — coastal coordinates

Fictional source situation: An intertidal survey point is mapped offshore after a datum mismatch.

Hidden failure: The translator explains the point as a marine transect without reviewing the coordinate reference system. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve geodeticDatum and escalate any reprojection as a geospatial data operation. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Reproject test points with documented parameters before writing a narrative. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

D10 — uncertainty unknown

Fictional source situation: A precise-looking old coordinate has no recorded uncertainty.

Hidden failure: The target infographic states ‘accurate to 1 metre’ based only on the number of digits. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Describe positional precision as unknown until a georeference audit supplies evidence. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check for empty uncertainty fields and avoid invented accuracy claims. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E01 — event year versus upload year

Fictional source situation: An observation from a past decade was uploaded to a portal this month.

Hidden failure: A target-language ‘new discovery’ headline uses the upload date as the biological event date. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Distinguish eventDate from publication and modification timestamps. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Verify the event date from the original record and keep the headline temporally honest. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E02 — estimated month

Fictional source situation: A curator interprets ‘summer’ in a historical label as a broad seasonal interval.

Hidden failure: The translation creates a specific calendar date for sorting convenience. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain verbatimEventDate and mark any normalized date as an interpretation with uncertainty. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check the inferred date against the source text and hemisphere context. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E03 — time zone ambiguity

Fictional source situation: A camera record gives local clock time without a documented UTC offset.

Hidden failure: The target interface appends Z, asserting the time was UTC. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep the original time basis and explain that its time zone is unspecified. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare device metadata before converting the timestamp. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E04 — trap-nights

Fictional source situation: A survey deployed five traps for four nights and describes 20 trap-nights of effort.

Hidden failure: A translated label says ’20 nights of sampling’. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve the effort unit and explain traps multiplied by nights. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check the multiplication and avoid mistaking effort for elapsed time. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E05 — transect length

Fictional source situation: A team walked 1.5 km per visit across repeated surveys.

Hidden failure: A summary translates 1.5 km as the entire monitoring distance for the season. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Separate per-visit effort from accumulated total and record the number of visits. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Recalculate the denominator for any encounter rate. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E06 — quadrat size

Fictional source situation: A vegetation study sampled 10 quadrats of 1 m² each.

Hidden failure: The target paragraph describes a ten-square-metre quadrat, changing the sampling geometry. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Translate quantity, size and replication separately: ten one-square-metre units. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check both total sampled area and spatial design. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E07 — nested event IDs

Fictional source situation: A single expedition includes several daily stations and multiple sampling events.

Hidden failure: The target reporting tool counts every station row as an independent expedition. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve parentEventID links and event hierarchy. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Group results by parent and child IDs to verify narrative totals. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E08 — repeated seasonal survey

Fictional source situation: A dataset has visits to the same plot in two seasons.

Hidden failure: The target map treats the repeated records as evidence of two distinct plots. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep locationID and eventID separate and state repeated visits explicitly. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check unique locations and unique events as different counts. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E09 — negative detection threshold

Fictional source situation: A laboratory result is reported as below detection limit, not equal to zero.

Hidden failure: The translated measurement table replaces it with 0. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve the less-than qualifier, unit and detection threshold. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Ask whether the analytical method could have detected the reported concentration. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

E10 — incomplete inventory

Fictional source situation: A checklist note says targeted sampling likely missed some cryptic taxa.

Hidden failure: The target asserts the species list is exhaustive. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Translate sampling completeness as a qualified estimate, not certainty. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Review targeted taxon scope, protocol and completeness commentary. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F01 — rare versus threatened

Fictional source situation: A report calls a moss locally uncommon without assigning a formal conservation category.

Hidden failure: The translated public label declares the species endangered. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep observation frequency and formal Red List status distinct. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Verify status in the referenced assessment authority and date. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F02 — regional assessment

Fictional source situation: A taxon has a regional conservation status not necessarily equal to a global assessment.

Hidden failure: The target press release labels the regional status as worldwide risk. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: State the assessment scale and authority wherever a category appears. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare the geographic scope of the two lists without assuming equivalence. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F03 — introduced plant

Fictional source situation: A field inventory records the presence of an introduced ornamental species.

Hidden failure: The translation uses ‘dangerous invasive species’ without documented impact. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve establishment status and make impact claims only with evidence. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check which authority defines the invasive label. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F04 — native at what scope

Fictional source situation: A species is native to one ecosystem but introduced within another range.

Hidden failure: A global brochure says the species is simply native everywhere. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Attach nativeness to the assessed range and documented classification. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Examine establishmentMeans and distribution context before generalising. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F05 — endemic wording

Fictional source situation: A species is restricted to a small region within a larger biological distribution system.

Hidden failure: The translator replaces endemic with ‘rare’ because the words sound interchangeable. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep the geographical exclusivity claim separate from abundance and threat status. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Ask for the exact scope supporting the endemic assertion. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F06 — rediscovery headline

Fictional source situation: A species known only from historical records is newly recorded at one sampling site.

Hidden failure: The translation says its global population has recovered. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Report renewed evidence of occurrence without inferring demographic trend. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Review whether repeated population monitoring exists. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F07 — ecological suitability model

Fictional source situation: A model identifies areas suitable for a taxon under assumptions and scenarios.

Hidden failure: The translated legend says those cells are confirmed occurrence locations. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Use ‘predicted suitable habitat’ rather than confirmed observations. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check modelling uncertainty and distinguish modeled pixels from sampled points. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F08 — habitat loss estimate

Fictional source situation: A study estimates projected habitat change under one scenario with confidence bounds.

Hidden failure: The target text presents the central estimate as an inevitable loss. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve conditional scenario language and ranges, including the projection time horizon. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Trace the baseline and scenario assumptions in the source. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F09 — environmental DNA

Fictional source situation: A water sample has eDNA reads assigned provisionally to a taxon.

Hidden failure: A media caption claims the organism was filmed in the water. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Distinguish molecular detection from direct visual observation and verified presence. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Review contamination controls and sequence classification quality. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

F10 — culturally restricted ecological record

Fictional source situation: A community-supplied habitat name has conditions on quotation and reuse.

Hidden failure: The translator treats it as freely reusable public vocabulary. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Respect the permission, community attribution and publication boundaries; use a consultation process. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare rights documentation with the intended output channel. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G01 — DwC header localised

Fictional source situation: A spreadsheet header occurrenceID is replaced with a translated phrase in a production export.

Hidden failure: Downstream applications fail because the parser expects the stable key. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep canonical headers in machine data and publish translated labels in a separate data dictionary. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Run a clean import against the agreed schema. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G02 — XML attribute altered

Fictional source situation: An archive metadata file uses machine-readable XML element names and language-dependent content values.

Hidden failure: The translator changes element names along with descriptions. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep schema syntax and identifiers stable; translate only authorised content-bearing nodes. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Validate XML against its schema after localisation. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G03 — ID leading zeros

Fictional source situation: A specimen code stored as text begins with two zeros.

Hidden failure: The translated spreadsheet opens it as a number and drops the zeros. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Protect identifier column types and quote text where required. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare original and exported IDs byte-for-byte. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G04 — Unicode diacritics

Fictional source situation: A collector’s name and taxon authorship include diacritics.

Hidden failure: A conversion step strips characters because the target font lacks glyph coverage. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Use suitable Unicode encoding and fonts without silently altering names. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Round-trip text from export through import and compare character sequences. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G05 — JSON key renamed

Fictional source situation: An API example contains key scientificName and a human-readable description beneath it.

Hidden failure: The translation localises the key but not the documentation’s example call. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep key scientificName unchanged; localise adjacent prose and field explanations. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Execute or validate a sample request using the translated documentation. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G06 — null overwritten

Fictional source situation: A blank recordedBy value is displayed as ‘anonymous observer’.

Hidden failure: The user interface invents a privacy claim not made by the source. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Use ‘not provided’ or another policy-approved missing-value display, without assigning a false author status. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Confirm database null semantics and access policy. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G07 — media licence omitted

Fictional source situation: An image caption is translated but its attribution and licence link disappear.

Hidden failure: The new page implies unrestricted ownership of the photograph. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Preserve rights information and link to the actual licence and attribution. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Perform media-rights review separately from scientific text QA. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G08 — translated term clashes

Fictional source situation: Two translators use ‘sample’ for one physical specimen and one environmental extract.

Hidden failure: A reader assumes both are the same evidence type. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Agree on separate glossary entries tied to basisOfRecord and materialSampleID. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Search all target assets for collisions and inspect context. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G09 — README stale

Fictional source situation: A translated readme describes an old column schema although the downloadable archive has changed.

Hidden failure: Readers follow instructions that no longer match the current dataset. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Bind the document version to a specific export and regenerate its field inventory when data change. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Diff source headers against the translated documentation before release. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

G10 — publication record versus citation

Fictional source situation: A dataset DOI identifies the dataset, while one study DOI cites research based on it.

Hidden failure: The translation uses the article link as if it were the dataset’s persistent source. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Distinguish dataset publication, paper and source observation references. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Resolve each DOI and check the metadata type. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H01 — legacy numeric taxonKey

Fictional source situation: A tutorial uses an integer key from the older GBIF Backbone as though it were universally current.

Hidden failure: Readers use the identifier in a default taxonomy interface and get an unexpected result. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Explain checklist context and the distinction between legacy and COL XR identifiers. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Test the sample against the correct API and version before publishing. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H02 — COL XR alphanumeric key

Fictional source situation: A current Catalogue of Life Extended Release key contains letters and numbers.

Hidden failure: A localisation routine insists that every taxonKey be numeric and strips letters. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Treat identifiers as strings and retain their taxonomy context. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Validate identifier type against the documented 2026 GBIF interface behaviour. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H03 — API default mismatch

Fictional source situation: A web interface and API endpoint use different default taxonomy behaviour during a migration.

Hidden failure: A translated help page assumes identical results for queries without checklistKey. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Document each endpoint and any required checklistKey explicitly. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Reproduce the same search in both interfaces to identify the difference. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H04 — fuzzy name match

Fictional source situation: An automated service reports a fuzzy match for a misspelled scientific name.

Hidden failure: The target narrative treats the match as proof that the source identified that accepted taxon. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Keep source spelling and matching flag; treat suggested matches as review candidates. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Inspect taxonomic match flags and confirm with an expert. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H05 — higher-rank match

Fictional source situation: A new species name cannot be matched exactly and the service resolves only to its genus.

Hidden failure: The translation invents a species-level accepted identifier. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: State the higher rank returned and preserve the unresolved original name. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Check match confidence and taxonRank before producing public claims. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H06 — synonym mapping

Fictional source situation: A historical scientificName is mapped to a current accepted usage in a registry.

Hidden failure: The translation discards the older string and loses literature searchability. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Retain both usage labels and the relationship with a cited authority. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Test discovery under both old and current names. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H07 — GBIF issue flags

Fictional source situation: An occurrence API response contains interpreted fields and quality issue indicators.

Hidden failure: The target display hides the issue flags to look cleaner. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Expose or explain relevant interpretation limitations without overstating data quality. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Compare source fields, interpreted fields and issue flags. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H08 — unreviewed machine translation

Fictional source situation: An automated translation turns a rare taxon epithet into an everyday adjective.

Hidden failure: The resulting sentence is fluent but nomenclaturally meaningless. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Protect scientific names with term locks and require qualified review of each record type. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Run exact-string QA and a bilingual domain evaluation. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H09 — mixed-language release

Fictional source situation: A collection publishes labels in several languages with no language identifiers.

Hidden failure: Users cannot tell whether two phrases are alternatives or different data claims. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Attach language tags to localised labels where the publishing system supports them. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Test selection, fallback and accessible display for each locale. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

H10 — false global comparison

Fictional source situation: Two portals use different taxonomic backbones and record filters, but an infographic compares their raw species totals.

Hidden failure: A translated caption calls the difference evidence of biodiversity loss. The problem is not merely awkward vocabulary; it changes what the reader would believe the source data established. Treat that difference as an evidence error even when every individual word seems plausible.

Safer translation decision: Explain taxonomy, spatial scope, time filter and deduplication before comparisons. Keep any formal identifier, numerical field, uncertainty marker or relationship in its assigned data layer, and use prose only to explain what the evidence actually warrants. Do not silently replace the source owner’s scientific judgment with an editorial guess.

Reviewer diagnostic: Re-run totals under aligned conditions or state why they cannot be compared. Ask a second reader to write down the implied object, event time, quantity, certainty and source authority after seeing only the target output. Compare that reconstruction against the original. A discrepancy should return to the relevant record owner or qualified reviewer, rather than being concealed by another stylistic pass.

Forty recurring failure modes and how to repair them

A translated common name is treated as a unique species ID

Failure mechanism: A common name can cover several taxa or differ by locale and community. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Use scientificName and taxon authority for identity; keep vernacular language and provenance separately. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Every empty field becomes zero

Failure mechanism: An absent value is not a measured zero and may indicate unavailable metadata. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Restore null values and define an honest missing-value display. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

The export date replaces eventDate

Failure mechanism: Publication tells when data became available, not when the organism was observed. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Verify eventDate and display update timestamps as a separate category. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Taxon names are silently modernised

Failure mechanism: An accepted-name interpretation is mistaken for the verbatim historical record. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve verbatim nomenclature and add explicit contemporary usage mapping. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

The record count becomes a population total

Failure mechanism: Duplicate events and media may represent the same organism. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Clarify which quantity is counted and inspect deduplication. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Negative observation becomes species absence

Failure mechanism: Failure to detect a taxon is not always affirmative evidence of absence. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Include sampling protocol and relevant detection limitations. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Coordinates are printed as exact sites

Failure mechanism: Displayed decimals convey more precision than the georeference permits. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Show uncertainty, generalisation and sensitivity policy. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A sensitive location is disclosed

Failure mechanism: Target-language detail inadvertently defeats deliberate location generalisation. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Review publication permission and omit restricted detail. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

ScientificName authorship becomes collector credit

Failure mechanism: Nomenclatural authors and recording agents serve different functions. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Map author, collector and identifier into separate fields. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A taxon split is imposed without authority

Failure mechanism: Source and receiving checklists may use different taxon concepts. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Record authority and date; preserve original classification. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A synonym is deleted from the record

Failure mechanism: Loss of historical names breaks research and specimen traceability. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Retain the synonym relationship and source spelling. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A specimen is described as a live observation

Failure mechanism: BasisOfRecord categories are confused. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Explain evidence class and avoid new claims of life or recency. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A tissue sample becomes an individual

Failure mechanism: MaterialSample is a physical or environmental sample, not automatically one animal. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Keep sample identity and sampling relationship separate. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

An uncertain determination becomes confirmed

Failure mechanism: Qualifiers and review status are smoothed away. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve uncertainty and determination history. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A local place name is aggressively normalised

Failure mechanism: A modern locality guess replaces a historical document. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve verbatimLocality with documented interpretation. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A map datum is inferred

Failure mechanism: Coordinate labels are treated as if every record uses the same datum. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Retain geodeticDatum and refer reprojection to a specialist. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Numeric formatting corrupts CSV

Failure mechanism: Locale separators conflict with delimiters and parser rules. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Test import/export and choose machine-stable numeric encoding. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Units disappear from a chart

Failure mechanism: Organism quantity, density and effort are incorrectly conflated. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Keep measurementValue, measurementType and measurementUnit together. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Evidence media loses attribution

Failure mechanism: A multilingual interface omits licensing and creator data. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Retain rights links and media provenance. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A machine field key is translated

Failure mechanism: Interoperability fails although human-facing labels appear improved. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Freeze canonical schema keys and localise display labels only. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A record is matched to the wrong kingdom

Failure mechanism: Ambiguous scientific names are resolved without higher classification. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Use taxon hierarchy and authority to disambiguate. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A fuzzy match is accepted as proof

Failure mechanism: An indexing algorithm is treated as a taxonomic expert. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Keep issue flags and request human scientific review. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Taxonomic keys are assumed numeric

Failure mechanism: Newer systems may employ alphanumeric identifiers. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Store taxonKey as a string with checklist context. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Different API defaults are ignored

Failure mechanism: A portal and search endpoint may return results organised under different taxonomies. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Document and test endpoint-specific query parameters. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Name authorship brackets are removed

Failure mechanism: Punctuation can encode nomenclatural history. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve complete name string and check approved formatting. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Field notes lose their original wording

Failure mechanism: Editorial improvement erases the evidence of historical transcription. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Provide both verbatim and interpreted forms. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

An image caption invents identification certainty

Failure mechanism: A photograph may lack sufficient diagnostic detail. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Link confidence to review evidence. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

An eDNA detection becomes a sighting

Failure mechanism: Genetic material does not necessarily prove a living individual at the exact coordinate. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Explain sampling and marker limitations. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

An acoustic count becomes individual count

Failure mechanism: Multiple calls can originate from one organism. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve detection units and counting protocol. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A conservation category loses its scale

Failure mechanism: Local and global assessment contexts are different. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Name authority, date and geographic scope. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A native species is called endemic

Failure mechanism: Native status does not imply exclusively restricted distribution. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Use precise distribution evidence before declaring endemism. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

An introduced species is labelled invasive

Failure mechanism: Introduction and demonstrated invasive impact are different claims. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve establishment status and sourced policy classification. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A provisional name is published as valid

Failure mechanism: Typography and fluency cannot confer nomenclatural validity. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Flag working names and require formal reference review. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Old and new dates are collapsed

Failure mechanism: Field, identification, publication and modification dates are distinct. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Write specific date labels. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Time zone is invented

Failure mechanism: A local timestamp is reformatted with an unsupported UTC marker. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve source time basis and document conversion. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Sampling effort is described as yield

Failure mechanism: Trap-nights or transect distance are mistaken for number of finds. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Explain effort unit and its statistical role. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

An incomplete inventory becomes complete

Failure mechanism: Limited coverage is rewritten as exhaustive species presence. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve target group, survey method and completeness qualifier. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A source interpretation is silently corrected

Failure mechanism: The translator makes scientific edits outside the approved scope. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Raise issues visibly and get the data owner’s decision. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A translated help file describes old schema

Failure mechanism: Documentation and dataset versions drift apart. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Bind help text to versioned headers and rerun structural QA. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

Cultural knowledge is treated as free data

Failure mechanism: Names and ecological statements may carry consent and community rules. A translation team can miss this because its ordinary grammar and fluency checks do not test the scientific relation carried by the statement.

Repair: Preserve attribution and honor permitted-use boundaries. Document the change request, preserve the earlier value in the evidence trail and run a small paired test showing what source claim and target claim now mean. When the answer depends on taxonomy, georeferencing, rights or data architecture, obtain review from the responsible specialist rather than relying on fluent paraphrase alone.

A release checklist: twenty-four independent quality gates

Gate: Record scope

Confirm which fields are translatable prose and which are schema, ID, controlled value, number or licence. Require a written classification for every edited column and any ambiguous field. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Identity preservation

Compare occurrenceID, eventID, organismID, materialSampleID and taxonID before and after localisation. Any changed persistent key needs an authorised migration, not a translator’s silent edit. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Row and link integrity

Check row counts, parent event relationships, foreign keys and linked media references. Reimport the translated package in a clean test environment. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Source traceability

Keep source dataset version, identifiers, original wording and revision logs accessible. A reader must be able to find the original scientific assertion. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Authority discipline

Name the checklist, assessment or nomenclatural treatment behind a scientific status claim. If authority is absent, the target should not sound more authoritative. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Term meaning

Compare every chosen display label with its current TDWG or domain definition. A bilingual expert should explain the concept without substituting a neighbouring field. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Scientific name control

Protect genus, epithet, rank, authorship, qualifier and original spelling. Do not accept general spell-check corrections as scientific validation. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Synonym relationship

Verify acceptedNameUsageID, originalNameUsageID and taxonomicStatus where used. A revision may add a current interpretation but must not erase the record’s history. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Evidence class

Check whether the record is observed, sampled, collected, inferred or cited. Reject any wording that upgrades evidence without new source support. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Temporal integrity

Check eventDate, verbatimEventDate, dateIdentified, modified date and download date separately. Make examples with clearly different dates to reveal conflation. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Spatial integrity

Compare signed coordinates, datum, coordinate uncertainty, altitude and depth. Plot test rows under the original and final display rules. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Sensitive locality protection

Confirm informationWithheld, dataGeneralizations and publication scope across languages. Run a disclosure test with a subject steward before public release. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Quantitative integrity

Validate values, units, significant digits, ranges, approximations and denominators. Use independent calculations for sample rates and effort. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Negative evidence

Distinguish no record, explicit absence, non-detection and below detection threshold. Require sampling context before writing absence conclusions. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Sampling design

Check samplingProtocol, samplingEffort, target taxon set and event hierarchy. Ask whether another team can reproduce the study design from the explanation. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Conservation language

Verify endemic, native, introduced, invasive, rare and threatened against stated evidence. Reject claims that upgrade a descriptive label to formal legal or conservation status. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Nomenclature currency

Compare taxonomic authority and version; inspect 2026 GBIF migration context when relevant. No assumption that legacy integer keys and new alphanumeric keys share format. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Machine compatibility

Validate CSV, XML, JSON, RDF or archive structures without depending on visual preview. A translated interface must not break parser keys. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Multilingual display

Check language tags, fallback behaviour, fonts, script direction, sorting and search. The record should remain identifiable when users switch languages. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Accessibility

Use understandable descriptions without erasing qualifiers, scientific precision or source roles. Ask target readers to paraphrase the record and compare to the source. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Rights and licensing

Check images, data licences, cited contributors, terms of reuse and policy restrictions. A language version must not silently grant permissions the original lacked. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Bilingual error review

Classify omissions, additions, wrong terminology, scope shifts and misleading certainty. Prioritise evidence-changing errors over cosmetic preferences. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Independent domain sign-off

Have an appropriate scientific or curation specialist review contested concepts and data corrections. Separate scientific acceptance from linguistic approval. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Gate: Release evidence

Archive checks, unresolved issues, final hash, data version and responsible sign-offs. A future maintainer should know exactly what was translated and how it was validated. Keep a short pass/fail record with the evidence used, the responsible reviewer and the version of the output under test. A checker who cannot identify the exact comparison should mark the gate unresolved, not silently pass it because the text is fluent.

Four-week implementation plan for an actual multilingual biodiversity project

Week 1 — Freeze and understand

Inventory source assets and publishing channels; verify canonical record types, data dictionaries, rights, authority choices and known gaps. Select a stable export or version and name the audiences. Work with curators and scientific reviewers to identify which statements are observations, interpretations and editorial explanations. Extract high-risk identifiers, names, dates, sampling effort and sensitive location fields. Build a change-control ledger before any translation begins.

Deliverable: one traceable source package, one reader-purpose brief, and a field classification sheet. Gate: a reader can tell what may be translated, what must remain immutable, and who can approve biological corrections.

A useful weekly review question is not whether the prose sounds impressive but whether a downstream reader could take the wrong scientific action because of a language decision. Revisit the cases that answer yes, rank their consequences and close the evidence gaps before widening publication.

Week 2 — Build terminology and prototypes

Create a small bilingual field dictionary with canonical keys, approved display labels and definitions. Prototype one specimen card, one occurrence table, one scientific-name summary and one geospatial display. Test common names, qualifiers, numeric units, identifiers, permission statements and date layers. Review the 2026 taxonomic reference and decide whether an accepted-name display is required or prohibited by the brief.

Deliverable: approved termbase, pilot interface, explicit taxon authority and a decision log for unresolved fields. Gate: independent reviewers can reconstruct exact source claims from four translated prototypes.

Week 3 — Translate and test records

Translate permitted descriptions and notes in controlled batches. Keep machine data frozen. Run string-lock checks on scientific names and identifiers, compare row counts and coordinate values, and perform sampling-based bilingual evaluations on cases with uncertain identification and complex histories. Ask curators to inspect suspicious scientific content and treat suggestions as tickets rather than silently fixing source data.

Deliverable: reviewed target-language assets plus structural test logs and recorded exceptions. Gate: the receiving application imports correctly and the explanatory text does not overstate evidence.

Week 4 — Publish, observe and maintain

Test real tasks with multilingual readers: find an occurrence, see its evidence type, read a specimen’s historical label, identify uncertainty and access its licence. Check that withheld locality remains withheld in every language. Publish with version labels, source links and a known-issues section. Establish triggers for future updates when Darwin Core terms, taxonomic backbones or original records change.

Deliverable: released localisation with reproducible QA records, sign-off and a maintenance plan. Gate: each translated claim can be traced to a stable source and responsible reviewer.

Transfer workshops: apply the same discipline beyond a species table

Museum specimen cards

Translate a cabinet card while preserving a faint historical inscription, its modern reading, an accession ID and a revised determination. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Ask two curators to recover original and interpreted content independently and explain why the two are not the same record. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Environmental impact report

Rewrite biodiversity baseline paragraphs for public readability while protecting dates, detection limitations and study area scope. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Give the draft to a technical reviewer to identify every place where survey absence has become ecological absence. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Citizen-science interface

Localise observation prompts that ask whether a user saw an organism, heard it or only found its track. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Test whether users can complete an observation without accidentally asserting identification certainty. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Species distribution map

Translate a map legend that includes occurrence, predicted habitat, historical record and withheld sensitive locality. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Ask readers to identify which icons are evidence, prediction and restricted data. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Educational glossary

Teach scientificName, vernacularName and taxonConceptID through progressively harder examples without promising one-to-one equivalence. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Test learners with a case where the same common name refers to different taxa. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Research abstract

Adapt an ecological study summary while preserving sample size, detection threshold, time window and uncertainty. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Compare claims line by line against the methods and results tables. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Data-portal API documentation

Translate help text while leaving JSON keys, URI patterns, example values and query parameters intact. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Run the tutorial requests and compare returned IDs with the source examples. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Herbarium digitisation

Use bilingual transcription notes to distinguish original handwriting from editorial interpretation of historical localities. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Audit a record where the locality is unresolved and must remain so. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Acoustic monitoring

Translate detection summaries so readers do not confuse call counts with population abundance. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Ask a domain reviewer to explain possible repeat calls and detector false positives. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Molecular survey

Write a plain-language eDNA description that says what genetic material suggests without inventing a witnessed living organism. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Check whether contamination controls, marker limits and confidence remain visible. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Taxonomy migration note

Explain why a legacy taxonKey differs from the current default classification identifier without silently replacing either. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Test old and new examples using the named checklist or API context. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Conservation briefing

Explain why rare, threatened, endemic and introduced name different questions, using attributed sources rather than impressions. This workshop is complete only when the chosen target language can do the reader’s real job without changing scientific identity, provenance or uncertainty. Make reviewers label each statement as observation, classification, risk assessment or inference. Record one example of an acceptable wording and one near-miss that sounds natural but asserts too much.

Frequently asked questions

Should Latin scientific names be translated?

Usually keep nomenclatural scientific-name strings intact, including relevant epithet, authorship, rank and qualifier. Translate nearby explanatory text or provide verified common names where suitable. A scientific name is a controlled identity-bearing expression, not ordinary prose. Context may require italic formatting for genus and species, but format alone does not validate nomenclature.

A practical check is to locate the exact source field or authority that supports the answer, then ask whether the target-language phrasing states more, less, or something different. If any substantive difference remains, explain the uncertainty rather than hiding it behind an elegant sentence.

Can I translate Darwin Core column names?

You can localise user-facing labels and documentation, but changing canonical machine-facing terms such as occurrenceID in an interoperable export is likely to break schema interpretation. Put translated captions in a data dictionary, interface label or secondary column when authorised; do not rewrite the normative key.

What is the difference between occurrenceID and organismID?

An occurrence identifies a record of evidence that a taxon occurred in a context; an organism identifier can refer to a particular organism followed across events. The two may have different multiplicities. Do not treat occurrence rows as a one-to-one tally of individuals.

What is the difference between eventID and occurrenceID?

An event generally describes sampling or observational activity, while occurrences record associated taxon evidence. One event may yield many occurrences and events can form hierarchies. Separate identifiers let analysts evaluate effort and observations without losing relationships.

Are specimen and sample interchangeable?

No. A specimen may be a preserved physical object in a collection. A material sample can be tissue, environmental material or another defined sampled object. Whether a record describes a specimen, sample or observation must come from its evidence fields and collection context, not editorial preference.

Can a translation update an outdated species name?

Only when the brief authorises an explicit modern interpretation and a qualified source supports it. Preserve historical nomenclature in the record and add accepted usage as a separately attributable mapping. A taxonomic change is a data interpretation, not routine word substitution.

What does accepted name mean?

An accepted name is accepted relative to a particular taxonomic treatment or reference checklist. It does not prove that every historical treatment or researcher used the same taxon concept. Name the authority and date when comparative interpretation depends on that distinction.

What is a synonym in taxonomy?

A synonym is a name usage linked under a particular treatment to a taxon with another accepted usage. It is not simply an everyday English synonym. Keep source and accepted usages and any relevant mapping keys, especially for older literature.

Can a common name have several scientific meanings?

Yes. Vernacular names vary across communities, languages and ranges and may refer to more than one taxon. Use scientific identity and source context to disambiguate, and treat common names as language-specific labels with provenance rather than universal keys.

What does cf. mean in a scientific name?

In many contexts cf. signals comparison with a named taxon and implies that definitive identification has not been asserted. Exact use varies by discipline. Do not delete the qualifier casually, and consult the source determination or subject specialist when in doubt.

Does ‘not observed’ mean absent?

Not necessarily. Non-detection depends on sampling design, effort, time and ability to detect the taxon. An explicit absence record can be meaningful under an established method, but no row in a database is not enough to assert ecological absence.

Does a GPS coordinate give the exact location?

No. It may be rounded, georeferenced from a place description, offset, generalized, or subject to a large coordinate uncertainty. Preserve geodetic datum and uncertainty fields and never infer accuracy solely from displayed decimal digits.

How should I translate old collection dates?

Keep the original writing and any normalized interpretation distinct. A vague year range or seasonal phrase should not be silently turned into a precise day. Date uncertainty is itself part of the specimen’s history and value.

How do I handle measurement units across languages?

Keep the scientific measured property, unit and significant distinctions attached to the value. Locale formatting may be adapted in display, but any conversion must be documented and independently checked. Never replace unavailable quantities with guessed values.

Are biodiversity data free to reuse?

Availability and reuse rights differ. Verify the record or media licence, source attribution requirements, exclusions and other restrictions. A translated page must not imply permissions broader than the actual source. Also review privacy and sensitive-locality controls.

What changed in GBIF taxonomy in 2026?

GBIF reported on 23 June 2026 that its website had changed its default organisation of occurrences to Catalogue of Life Extended Release. At the time described, API endpoints still required attention to checklistKey behaviour, and older GBIF Backbone integers could differ from newer alphanumeric keys. Consult current documentation for the exact endpoint in use.

How should machine translation be used?

Machine or AI output can accelerate first drafts, but protect schema keys and scientific names, restrict unsupported claims and require comparison against the source. A fluent hallucination is a severe error in a biodiversity record; qualified domain review is essential for disputed taxonomy or evidence interpretation.

Does ISO 5060:2024 guarantee a correct biodiversity translation?

No. ISO 5060 provides general guidance for evaluating translation output, including error types and quality ratings. Specialist scientific validation, source data governance, rights review and software testing are additional project controls. Do not mistake a general quality method for certification of scientific facts.

What about indigenous or community ecological knowledge?

Names, categories and ecological relations can belong to a community context with permissions and attribution expectations. Preserve provenance, avoid presenting a local category as a universal scientific equivalent and follow appropriate consultation and consent before broad republication.

What if the source record is scientifically wrong?

A translator should flag the suspected error, document evidence, and request the authorised owner or specialist to decide. Silently repairing the original data during translation loses provenance. A correction, when approved, should have its own change history.

Should I translate licence codes such as CC BY?

Keep the licence identifier and link intact and translate the user-facing rights explanation accurately. Attribution and redistribution constraints should remain prominent. Do not invent a different licence, assume universal public domain or remove contributor credit.

How do I audit a multilingual species map?

Test spatial coordinates, uncertainty, legend categories, date filters, taxonomic reference and sensitive-locality controls across language views. Ask a reader to explain what each symbol proves. A beautiful map is wrong if it turns historical observations or model predictions into current confirmed range.

How do I preserve identification history?

Keep each determination’s contributor, date, scientific-name usage and review status where the data model supports it. Present a current view without erasing older interpretations. The history can be essential for tracing studies based on earlier classifications.

What is the fastest safe way to begin?

Choose one canonical export and one small representative subset: a historic specimen, a citizen observation, a sampling event, an uncertain name, a sensitive coordinate and a modern synonym. Translate permitted prose, lock structural fields, and run scientific and bilingual review before scaling.

Terminology ledger: a compact field-by-field reference

Darwin Core

Darwin Core: TDWG vocabulary for interoperable biodiversity information; treat terms as a controlled reference, not arbitrary prose. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

DwC-A

DwC-A: Darwin Core Archive packaging convention for publishing structured biodiversity records and metadata. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

GBIF

GBIF: Global Biodiversity Information Facility, a discovery and data network with indexing interpretations. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

IPT

IPT: Integrated Publishing Toolkit used to prepare and register biodiversity datasets. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Taxon

Taxon: Biological group as delimited by an identified taxonomic treatment. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Taxon concept

Taxon concept: Taxon circumscription under a specific authority, not simply a written name. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

taxonID

taxonID: Identifier for a taxon within an appropriate data context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

taxonConceptID

taxonConceptID: Identifier for a circumscribed taxon concept under a reference. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

scientificName

scientificName: Scientific taxonomic name string associated with the source assertion. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

scientificNameID

scientificNameID: Identifier for the scientific name rather than every use or concept. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

scientificNameAuthorship

scientificNameAuthorship: Nomenclatural authorship associated with the name. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

taxonRank

taxonRank: Rank of the most specific name in a taxonomic hierarchy. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

acceptedNameUsageID

acceptedNameUsageID: Reference to the accepted usage in the chosen nomenclatural or taxonomic treatment. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

originalNameUsageID

originalNameUsageID: Reference to the original usage in a nomenclatural relationship. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

taxonomicStatus

taxonomicStatus: The status of a name usage under the described classification. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

vernacularName

vernacularName: Everyday name whose language, community and range matter. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

nameAccordingTo

nameAccordingTo: Reference indicating the treatment according to which a name usage is defined. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

namePublishedIn

namePublishedIn: Publication information for a name, not a sighting source by default. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

genus

genus: Taxonomic rank and component of many binomial names. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

specificEpithet

specificEpithet: Name component used with a genus in species nomenclature. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

infraspecificEpithet

infraspecificEpithet: Name component below species where applicable. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

identificationID

identificationID: Identifier for a determination or identification act. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

identificationQualifier

identificationQualifier: Marker qualifying the identification and its certainty or relationship. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

identificationVerificationStatus

identificationVerificationStatus: Recorded status of an identification’s review or verification. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

identifiedBy

identifiedBy: Person or group assigning a biological identification. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

dateIdentified

dateIdentified: Date of identification, which may differ from the collection date. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Occurrence

Occurrence: Evidence that a taxon occurred in an event or context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

occurrenceID

occurrenceID: Identifier for an Occurrence record, ideally persistent and unambiguous. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

occurrenceStatus

occurrenceStatus: Declared presence or absence when supplied by source data. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

basisOfRecord

basisOfRecord: Controlled indication of the nature of the recorded evidence. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Organism

Organism: A biological individual or associated organism concept in the source model. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

organismID

organismID: Identifier connecting records relating to a particular organism. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

organismQuantity

organismQuantity: Reported quantity associated with organisms in a source record. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

organismQuantityType

organismQuantityType: The kind of quantity, such as count, biomass or coverage. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

individualCount

individualCount: A count field whose meaning should not be inferred from row number. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Event

Event: Sampling or observational activity connecting records to protocol, place and time. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

eventID

eventID: Identifier connecting event and associated occurrences. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

parentEventID

parentEventID: Identifier representing event hierarchy, when used. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

eventDate

eventDate: Date or interval associated with the event. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

verbatimEventDate

verbatimEventDate: Original event date wording before normalized interpretation. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

samplingProtocol

samplingProtocol: Description or identifier of how a field event gathered evidence. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

samplingEffort

samplingEffort: Amount of effort applied to the sampling protocol. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

sampleSizeValue

sampleSizeValue: Magnitude of the sample size measure, not an implicit organism count. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

sampleSizeUnit

sampleSizeUnit: Unit attached to a sample size measure. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

MaterialSample

MaterialSample: Physical or environmental material obtained through sampling. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

materialSampleID

materialSampleID: Identifier for a material sample. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

catalogNumber

catalogNumber: Curatorial catalogue number that may only be locally unique. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

institutionCode

institutionCode: Code identifying the institution in context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

collectionCode

collectionCode: Code distinguishing an institutional collection in context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

recordedBy

recordedBy: Person or group associated with recording or collecting the occurrence. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

recordNumber

recordNumber: Identifier from the collector’s numbering system where present. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

otherCatalogNumbers

otherCatalogNumbers: Related local curatorial numbers, not automatic global keys. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

associatedMedia

associatedMedia: References to media connected to an occurrence or evidence object. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

associatedReferences

associatedReferences: References relevant to an occurrence or related assertion. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

locality

locality: Specific description of the place associated with the record. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

verbatimLocality

verbatimLocality: Original locality string recorded before interpretation. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

countryCode

countryCode: Country code in a controlled coding system when applicable. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

waterBody

waterBody: Name or identification of a body of water in the record. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

habitat

habitat: Description of environmental context, not a geocoded identifier. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

locationID

locationID: Identifier linking records to a location concept. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

decimalLatitude

decimalLatitude: Numerical latitude coordinate in the declared geographic reference context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

decimalLongitude

decimalLongitude: Numerical longitude coordinate in the declared geographic reference context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

geodeticDatum

geodeticDatum: Reference datum or coordinate system associated with coordinates. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

coordinateUncertaintyInMeters

coordinateUncertaintyInMeters: Spatial uncertainty of coordinates as described by the standard field. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

coordinatePrecision

coordinatePrecision: Precision of original coordinate representation, distinct from accuracy. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

georeferenceProtocol

georeferenceProtocol: Documented method used to assign a geographic reference. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

georeferencedBy

georeferencedBy: Person or group responsible for the assigned georeference. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

georeferenceSources

georeferenceSources: Evidence sources supporting the georeference. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

minimumElevationInMeters

minimumElevationInMeters: Lower elevation bound when recorded. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

maximumElevationInMeters

maximumElevationInMeters: Upper elevation bound when recorded. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

minimumDepthInMeters

minimumDepthInMeters: Lower depth measure when recorded. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

maximumDepthInMeters

maximumDepthInMeters: Upper depth measure when recorded. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

sex

sex: Biological field describing the recorded organism where applicable. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

lifeStage

lifeStage: Recorded developmental stage of the organism. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

reproductiveCondition

reproductiveCondition: Recorded condition relating to reproduction. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

behavior

behavior: Observed behaviour or evidence type where documented. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

establishmentMeans

establishmentMeans: Information about how a taxon came to occur in an area. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

degreeOfEstablishment

degreeOfEstablishment: Information about establishment stage under a defined vocabulary. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

pathway

pathway: Information about introduction pathway when relevant. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

measurementType

measurementType: Name of measured or assessed property in a measurement record. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

measurementValue

measurementValue: Reported measurement value. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

measurementUnit

measurementUnit: Unit needed to interpret a measurement value correctly. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

measurementAccuracy

measurementAccuracy: Accuracy or uncertainty statement attached to measurement context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

informationWithheld

informationWithheld: Explanation of information deliberately excluded from public record. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

dataGeneralizations

dataGeneralizations: Description of how data were generalized, including sensitive localities. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

license

license: Machine-readable or cited permissions for data reuse. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

rightsHolder

rightsHolder: Entity holding relevant rights in the content. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

bibliographicCitation

bibliographicCitation: Recommended bibliographic citation for a data resource. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

datasetID

datasetID: Identifier for the dataset whose records are being described. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

datasetName

datasetName: Human-readable name of the containing dataset. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

modified

modified: Timestamp of resource modification; not necessarily eventDate. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

language

language: Information about the language of source content or a label. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Humboldt Extension

Humboldt Extension: Inventory and survey-completeness vocabulary incorporated into Darwin Core. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

taxonKey

taxonKey: Taxonomic indexing identifier whose meaning depends on named checklist and API context. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

checklistKey

checklistKey: Parameter selecting a taxonomic checklist in relevant GBIF service calls. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

COL XR

COL XR: Catalogue of Life Extended Release, used for newer GBIF default website taxonomy in 2026. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

GBIF Backbone

GBIF Backbone: Older taxonomic backbone retained in some matching and API contexts. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

fuzzy match

fuzzy match: Approximate nomenclatural match requiring review, not proof of source identity. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

higher-rank match

higher-rank match: Taxonomic resolution above the claimed target rank when exact match fails. For translation, check the source’s canonical definition, whether this is an identifier, a machine-controlled value or a human-readable field, and whether a target-language label would change its evidential meaning. A term’s visual familiarity is not proof that its semantics match an ordinary-language near-synonym.

Final diagnostics: ten questions before publication

Could the translation make one individual appear as several organisms, or several observations appear as an independent population?
Can a reader distinguish what was observed from how the observation was later identified?
Are historical labels, accepted names, synonyms and chosen checklist versions simultaneously traceable?
Have all uncertain, approximate, negative and missing values retained their different meanings?
Do dates distinguish field event, identification, digitisation and publication?
Are sampling effort, method, quantity type and units visible where conclusions depend on them?
Are coordinates, uncertainty, reference datum and sensitive-locality controls preserved?
Can software still read every identifier, schema key, separator and linked reference as intended?
Have data rights, media licences and community permissions survived into every target-language channel?
Can a qualified reviewer reconstruct every substantive public claim from the source and named authority?

Next routes: a system of owners, not competing encyclopaedias

Return to the Master Art of Translation | The Complete System for Moving Meaning Between Languages for the underlying meaning-preservation architecture, source analysis, terminology, machine assistance, revision and transfer. This child specialises only in biodiversity records, taxonomic interpretation and scientific-evidence language. It should not become a generic definition page for every ecological topic or a substitute for the original scientific authority.

When the missing ability is learning technical words rather than translating a structured biodiversity claim, continue to the protected eduKateSG Vocabulary Learning Hub. When the problem is English grammar, inference, sentence clarity or scientific writing inside English, use the protected How English Works | Translation and Meaning Across Languages route. These owners supply adjacent skills without being rewritten or duplicated here.

To verify source concepts, visit TDWG Darwin Core List of Terms (version 13 October 2026), TDWG Darwin Core standard, GBIF IPT publishing guide, GBIF taxonomy interpretation, GBIF migration guidance (23 June 2026), TDWG Humboldt Extension and ISO 5060:2024. The publication date of this chapter does not make these external references immutable; refer to the current edition before applying a rule operationally.

The lasting principle is uncomplicated: when a biodiversity statement crosses languages, its words may change, but its source identity, evidence class, taxonomic interpretation, time, space, measurements, uncertainty, provenance, and permissions must not be silently transformed. Translation succeeds when a new reader gains access to the same justified meaning rather than a more confident story than the data can support.

Master Art of Translation | The Biodiversity Records, Taxonomic Names & Species Occurrence Translation System — How Darwin Core, GBIF, Specimens and Conservation Evidence Keep Their Meaning