Master Art of Translation | The Genebank Accession, MCPD Passport Data & Plant Genetic Resources Translation System — How DOIs, Taxonomy, Provenance and Availability Keep Their Meaning

The answer: how to translate genebank accession data without changing the specimen, provenance or scientific claim

How do you translate genebank accession records, MCPD passport data, plant genetic resources descriptions and seed-bank catalogues accurately? Start by treating every record as a scientific assertion about a particular accession rather than as a paragraph of ordinary prose. Protect the FAO/Bioversity Multi-Crop Passport Descriptors (MCPD) field names, the holding institute code, accession number, botanical taxon, collecting evidence, dates, coordinates, status flags and persistent identifiers. Translate the human-readable explanation; do not silently translate or reconstruct the structured identifiers. This guide explains a practical workflow for genebank passport data translation, multilingual germplasm databases, accession metadata quality assurance and cross-language records that remain useful to researchers, curators and people seeking plant genetic resources.

The hidden problem is that a fluent translation can change the identity of the biological material. A genebank accession number is not a collecting number, a donor accession number is not the present holder’s number, and the country where material was originally collected is not necessarily the country where its seeds are stored. A DOI assigned through the Global Information System for plant genetic resources is a persistent reference, not a replacement for every local accession identifier. The same record may distinguish a wild relative, landrace, breeding line, field collection, cryopreserved sample, physical safety duplicate and historical catalogue entry. Translators who collapse those distinctions create searchable text that looks correct while breaking research links or overstating access to material.

To translate MCPD 2.1 records correctly, first map the source fields to their authoritative meanings, then preserve machine-readable codes and values, and finally revise narrative fields in the target language with a traceable crosswalk. The correct result is not merely understandable: it must still identify the same accession, reflect the same evidence and allow a curator to detect what is unknown. The method below covers FAO WIEWS institute identifiers, Genesys PGR accession records, ITPGRFA/GLIS digital object identifiers, collection-site uncertainty, biological status codes, safety-duplication claims, material requests and multilingual documentation. It is world-facing and independent of any one country or language pair.

Architecture: EDKSG-TRANS-MASTER-WORLD-830. Purpose: an educational, operational translation handbook, not legal advice, a substitute for an institute’s accession-management policy, or an instruction to publish sensitive collecting locations. For the general translation framework, see Master Art of Translation | The Complete System for Moving Meaning Between Languages. The companion Vocabulary Learning Hub supports terminology learning; How English Works | Translation and Meaning Across Languages supports sentence-level explanation and audience adaptation.

Quick-start control sheet: preserve identity before you improve the wording

Establish the record’s home: which holding institute, collection and accession is actually meant? Never begin with a free-text crop name as the sole key.
Lock structured values: field identifiers, institute codes, local accession numbers, DOI strings, dates, controlled codes, original coordinates and source quotations.
Separate observations from interpretations: do not change a reported origin, sample status or provenance pathway to match what seems botanically probable.
Keep uncertainty visible: an unknown month is not January; a generalized coordinate is not a precise GPS fix; a blank distribution flag is not refusal.
Distinguish conservation, eligibility and distribution: presence in a genebank, legal access status and physical stock for dispatch are different facts.
Deliver two layers: machine-readable values unchanged and a human-readable explanation that can be reviewed by a subject specialist.
Validate before release: stable row count, field names, joins, missing values, codebooks, spelling of scientific names, sensitive-site review and paired-language comparison.

What this article covers, and what it deliberately does not replace

This handbook concerns the translation of accession-level passport and related curatorial metadata. It is not a seed certification guide: a lot number, germination certificate, varietal purity claim or phytosanitary certificate may accompany a seed shipment, but those records answer different questions from an accession’s conservation and provenance history. It is not a botanical-taxonomy authority either. A translator can preserve the taxon asserted by the source and flag an apparent synonym for review, yet cannot decide an accession’s definitive identification from wording alone. Nor does a multilingual display label confer rights to distribute genetic resources. Those decisions belong to authorized collection managers and applicable agreements.

An accession is a managed unit of germplasm within a collection, with an identity and history. Material derived from that accession may change through regeneration, selection, sampling, distribution or transfer. A sample is not necessarily the same conceptual object as an accession record; a donor designation is not automatically its current holding designation; and a DOI is a reference to a defined plant genetic resource, not proof that every related seed lot is identical. Preserve these levels explicitly in all target languages. If a source system uses a broader or narrower definition, record that scope rather than hiding it behind familiar terminology.

The FAO/Bioversity MCPD 2.1 reference remains a widely used field model for the exchange of accession passport information. The operational Genesys documentation supplies additional provider-side rules, including minimum identifiers, validation and service-specific fields such as AVAILABLE, HISTORIC and CURATION. In July 2026 Genesys published updated API guides distinguishing bulk accession passport retrieval from trait-data retrieval. A translator must not mistake a changing software presentation or API mapping for a change in the scientific identity of the accession. Source: MCPD 2.1 descriptor reference, Genesys passport-data API guide.

The hidden problem: translation crosses five identities, not one

1. Accession identity

A local accession number exists within the holding institute’s namespace; it is not necessarily globally unique alone. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Compare a local number with an institute code before merging two records. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. The same visible number at two different institutes may designate entirely different germplasm. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

Apply an adversarial test: give the source and target versions to different readers, and ask each to identify the holding accession, relevant actor, evidence date, possible requestability and unresolved uncertainty. Any disagreement is a translation defect to investigate, not merely a stylistic preference. Record the field, proposed correction, reviewer and supporting record before release.

2. Collection history

The collector, collecting mission, original locality, donor, receiving institution and subsequent safety-duplication site may all differ. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Represent each actor in the role assigned by the record. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. A provenance statement becomes false when the translator makes the latest holder the original collector. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

3. Taxonomic assertion

Genus, species, authority, subtaxon and common crop name serve distinct purposes. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Retain scientific-name components and identify which name is a curatorial assertion versus a verified current match. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. A common-language crop name can cover multiple taxa, while a scientific taxon may have familiar synonyms. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

4. Time and spatial evidence

Dates may have incomplete precision; site coordinates may have known uncertainty or refer to a later georeferencing action. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Check the literal value, precision, datum, hemisphere, method and uncertainty together. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. A complete-looking date or location can create unjustified scientific certainty. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

5. Regulatory and access state

MLS participation, biological status, availability for distribution, historic state and conservation storage are independent fields. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Use source-defined code meanings and identify the responsible decision maker. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. A reader should not infer that ‘in storage’ means ‘available to order’ or that a DOI means permission to distribute. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

6. Record and material state

A published passport row, physical seed lot, archival record, stored DNA sample and requestable accession are related but not interchangeable. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Keep record-level and material-level statements visibly separate. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. Someone can find metadata about a historic accession that is no longer physically maintained. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

7. Linguistic state

A bilingual description can clarify provenance without changing a machine key. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Preserve original value plus translated display value with language metadata. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. Replacing original prose with target prose can destroy the audit trail even when the summary seems adequate. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

8. Evidence state

A fact can be sourced, inferred, disputed, obsolete, corrected or merely unreported. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Attach provenance and review state to corrections instead of overwriting silently. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. An apparently impossible collecting site may reflect a georeferencing problem, a historical boundary or a mistaken transposition. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

9. Network state

The same accession can appear in local systems, Genesys, GLIS and publications through different identifiers and links. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Maintain a one-to-many identifier crosswalk with link provenance. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. Cross-portal search quality suffers when aliases are treated as new accessions. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

10. Ethical state

Exact locality descriptions and collector details can implicate fragile plant populations, communities and contractual restrictions. The practical consequence for a translator is that the unit of meaning is not an isolated word but a typed relationship: who designated the material, what type of material it is, when the observation was made, and with what authority. A useful preliminary annotation names the record owner, identifies the source field and notes the possible competing interpretation. That annotation should remain outside the published value unless the source itself supports it.

Apply disclosure review separately from translation accuracy review. Ask whether the proposed target wording would allow two independent curators to select the same underlying record. A perfectly literal translated collecting site may still be unsuitable for public release. When the answer is no, preserve the source value visibly and add a cautious explanatory note rather than rewriting the value into a more familiar phrase. A genuinely faithful translation protects what the record proves and refuses to supply the missing evidence.

The repeatable method: a 28-gate workflow from source intake to release

Gate 01 — Define the translation job

Identify language pair, audience, purpose and whether the output is an API payload, spreadsheet, catalogue page, sample request or explanatory note. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Do not allow a website translation to silently become an authoritative database migration. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 02 — Name the holding collection

Locate the collection owner and the source snapshot date, including the correct institute code if one is supplied. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A similar name can refer to a holding genebank, donor or collecting organization. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 03 — Freeze the original

Preserve a read-only source copy with version, export time, field order, row count and checksum if available. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Without an immutable baseline it is impossible to distinguish a translation correction from a new data update. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 04 — Declare the schema

Record whether data is MCPD 2.1, Genesys extensions, a local schema or an export crosswalk. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

An identically named column can have different value constraints in a different schema. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 05 — Inventory identifiers

Separate DOI/PUID, INSTCODE, ACCENUMB, UUID, COLLNUMB, DONORNUMB and OTHERNUMB. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Never combine identifiers merely because they look similar. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 06 — Test key uniqueness

Check composite accession keys in the source and in the translation export. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Deduplication by accession number alone can erase valid holdings. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 07 — Set an immutable token rule

Lock codes, DOI strings, geographic coordinates, dates, numeric units and controlled-value integers. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Translate labels around tokens, not the tokens themselves. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 08 — Map translatable fields

Tag free-text notes, collecting site descriptions and human-readable organization names separately from codes. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Do not insert translated prose into a machine enum field. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 09 — Declare a terminology authority

For each technical term cite a source codebook, authoritative manual or controlled vocabulary. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A search result is not automatically an approved taxonomy or legal interpretation. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 10 — Build field-level risk tiers

Prioritize identity, access, legal, coordinates, taxonomy and time before cosmetic style. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Review effort should follow potential consequences, not passage length. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 11 — Check specimen granularity

Determine whether the statement addresses accession, sample, lot, collection or taxon. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A stock-status sentence can be true for one lot but false as a claim about all material. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 12 — Preserve collecting vs acquisition

Distinguish the date collected from the date received or accessioned. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Confusing them can invent time in conservation or ownership history. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 13 — Check provenance actors

Label collecting, breeding, donor, holding and duplicate institutions with their separate roles. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

An actor in one role is not automatically the source of every observation. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 14 — Check country code semantics

Keep origin code apart from holding institute location and source language. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A translated country name must not replace the source-defined ISO alpha-3 code. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 15 — Validate coordinates

Check decimal sign, DMS conversion, datum, precision, altitude and uncertainty. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A sign change can move the locality across hemispheres. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 16 — Screen sensitive locations

Review endangered taxa, community information, embargoes and source rights before public release. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

An accurate translation can still expose information not approved for publication. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 17 — Check taxon hierarchy

Parse GENUS, SPECIES, SPAUTHOR, SUBTAXA and SUBTAUTHOR separately. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Do not turn a species authority abbreviation into a cultivar name. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 18 — Check status enums

Verify SAMPSTAT, COLLSRC, STORAGE, MLSSTAT and Genesys extensions against the current source scheme. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Never assume a code means the same thing across all schemas. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 19 — Distinguish blank and false

Represent missing, unknown, inapplicable and negative values according to actual system rules. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A blank AVAILABLE is not automatically unavailable. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 20 — Resolve multilingual names

Use source forms and language-tagged display aliases for crop and accession names where appropriate. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Never overwrite the controlled identifier with the local-language nickname. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 21 — Preserve relationships

Maintain references to donor, collector, mission and duplicate records with directional links. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A single translated ‘from’ can invert a donor–recipient relationship. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 22 — Check technology artifacts

Protect CSV delimiters, UTF-8 characters, JSON types, spreadsheet text formatting and API field names. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Leading zeros and literal plus signs may be lost before a reviewer sees the language. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 23 — Review evidence claims

Mark inferred taxonomy, inferred site, guessed year or expected availability as unverified. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A plausible deduction is not a source fact. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 24 — Perform bilingual QA

Have a language reviewer test meaning and a subject specialist test domain relationships. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

One reviewer alone may catch grammar while missing a wrong identifier role. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 25 — Run structural QA

Compare input/output field inventory, row counts, keys, missingness, code set and link integrity. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

A perfect sentence is irrelevant if the corresponding row disappeared. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 26 — Test a real retrieval path

Try searching and joining by the identifiers likely to be used by downstream systems. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Correct display alone cannot prove database interoperability. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 27 — Publish with version trail

Document source, date, glossary version, changes, exceptions and contact for corrections. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Do not announce a historical export as live current stock status. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Gate 28 — Review drift

Recheck when the collection record, source codebook, taxonomy or request process is updated. This gate has a concrete deliverable: a recorded decision and a reviewer who can reproduce it. The deliverable may be an identifier crosswalk, approved field map, source citation, privacy ruling, taxonomy note, structural comparison or signed release entry. Record not only what was changed but which values were deliberately left alone. Silence about a protected field can otherwise be misread as accidental omission.

Translation quality is maintained through change control, not one final proofreading pass. A useful negative test deliberately substitutes a plausible but wrong value into a copy of the record and asks whether the review would catch it. The test must use fictional data and must never modify a live accession record. If the review cannot distinguish the false statement from the supported source, strengthen this gate before proceeding. For a multilingual release, inspect both the rendered page and the underlying machine export; they can fail differently.

Descriptor-by-descriptor translation: the MCPD 2.1 semantic crosswalk

The following entries separate official field semantics from explanatory translation judgments. Each example value is illustrative unless explicitly linked to an authoritative record. This is a learning crosswalk, not a substitute for the complete descriptor reference and its coding notes. Use the current specification and provider validation rules to determine accepted formats. A field name in code type is a machine-readable label and should not be translated inside the data file; a human-facing caption for the field can be localized.

1. PUID — meaning, non-equivalence and review

What the record says. PUID refers to persistent unique identifier assigned to plant genetic resource material; Genesys accepts a DOI here. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Replacing a DOI with a catalog URL or arbitrary UUID. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Retain the exact DOI string and verify its resolver. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are GLIS and local accession references. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

Diagnostic test. Give one reviewer only the source cell plus column heading and another only the translated rendering. Ask both to describe the referent in one sentence, identify which information is fixed versus uncertain, and name the closest field that must not be confused with it. A disagreement reveals a concrete semantic defect. If the record is intended for automated ingestion, test its exact field code and type with the destination validator; back-translation alone cannot detect broken field names or code values. Store the issue as a named testcase so that subsequent language pairs inherit the correction.

2. INSTCODE — meaning, non-equivalence and review

What the record says. INSTCODE refers to wIEWS institute code for the institute holding the accession. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Mistaking the donor or collecting institution for the current holder. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve the registered code and separately translate the organization name. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are ACCENUMB and COLLNUMB. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

3. ACCENUMB — meaning, non-equivalence and review

What the record says. ACCENUMB refers to identifier assigned to the accession within the holding genebank. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Removing spaces, prefixes or leading zeroes during translation. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Treat as a literal string, not an ordinary number. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are Composite key with INSTCODE and GENUS. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

4. COLLNUMB — meaning, non-equivalence and review

What the record says. COLLNUMB refers to original collecting number assigned by collector or mission. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Renaming it as the genebank accession number. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep the original collector-level identifier and its relation to other holdings. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLMISSID, DONORNUMB. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

5. COLLCODE — meaning, non-equivalence and review

What the record says. COLLCODE refers to wIEWS code of the institute that collected the sample. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Assuming the collecting institute is necessarily the current holder. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep role and code distinct from INSTCODE. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLNAME when no code exists. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

6. COLLNAME — meaning, non-equivalence and review

What the record says. COLLNAME refers to collecting institute name used when a WIEWS code is not available. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Inventing an official institute code from a translated name. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Translate for readers but preserve the original name as an evidence string. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLCODE and COLLINSTADDRESS. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

7. COLLINSTADDRESS — meaning, non-equivalence and review

What the record says. COLLINSTADDRESS refers to address of collecting institute when its registered code is unavailable. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Turning a historic address into a present registered location. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Retain provenance, language and period where known. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLNAME and COLLCODE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

8. COLLMISSID — meaning, non-equivalence and review

What the record says. COLLMISSID refers to identifier used for the collecting mission. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Treating mission identifier as plant population identifier. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep exact characters and link to mission documentation. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLNUMB and COLLDATE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

9. GENUS — meaning, non-equivalence and review

What the record says. GENUS refers to botanical genus component of taxonomic name. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Replacing scientific genus with the ordinary crop word. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Retain genus capitalization and source attribution. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are SPECIES and CROPNAME. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

10. SPECIES — meaning, non-equivalence and review

What the record says. SPECIES refers to specific epithet, generally lowercase; sp. is an allowed uncertainty marker. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Capitalizing as a common name or asserting a species not identified. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep precise epithet or sp. source state. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are GENUS and SPAUTHOR. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

11. SPAUTHOR — meaning, non-equivalence and review

What the record says. SPAUTHOR refers to botanical authority of species name. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Mistaking taxonomic authority for the collector or an author of the article. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep scholarly authority notation without translating abbreviation. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are GENUS and SPECIES. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

12. SUBTAXA — meaning, non-equivalence and review

What the record says. SUBTAXA refers to subspecific or other additional taxonomic designation. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Confusing botanical variety with commercial cultivar or translating var. as variety brand. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Retain rank abbreviation and taxon text. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are SUBTAUTHOR and ACCENAME. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

13. SUBTAUTHOR — meaning, non-equivalence and review

What the record says. SUBTAUTHOR refers to authority for the most detailed additional taxon. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Deleting authority as nonessential punctuation. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Protect naming attribution and parse it separately. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are SUBTAXA and SPAUTHOR. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

14. CROPNAME — meaning, non-equivalence and review

What the record says. CROPNAME refers to common crop label that can vary across languages. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Treating a translated common name as definitive taxonomy. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Translate display label while keeping botanical identity fields. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are GENUS and SPECIES. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

15. ACCENAME — meaning, non-equivalence and review

What the record says. ACCENAME refers to accession names or designations, possibly multiple. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Merging synonyms into a new cultivar or dropping semicolon-separated aliases. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve source names and add language-tagged explanations. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DONORNUMB and COLLNUMB. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

16. ACQDATE — meaning, non-equivalence and review

What the record says. ACQDATE refers to date material entered the holding collection. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Equating acquisition with original collecting date. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve incomplete YYYYMMDD precision if source incomplete. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLDATE and origin history. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

17. ORIGCTY — meaning, non-equivalence and review

What the record says. ORIGCTY refers to three-letter country code for original collection, breeding or selection. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Replacing origin with the current holding institution’s country. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep ISO alpha-3 code and treat historical interpretation cautiously. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLSITE and SAMPSTAT. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

18. COLLSITE — meaning, non-equivalence and review

What the record says. COLLSITE refers to textual description below country level of original collecting place. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Over-precision or unsafe revelation of sensitive site. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Translate description without adding exact coordinates. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DECLATITUDE and COORDUNCERT. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

19. DECLATITUDE — meaning, non-equivalence and review

What the record says. DECLATITUDE refers to latitude of collecting site in decimal degrees, positive north. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Dropping minus sign or swapping comma decimal conventions. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve sign and numeric precision. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DECLONGITUDE and COORDDATUM. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

20. LATITUDE — meaning, non-equivalence and review

What the record says. LATITUDE refers to alternative latitude representation using DMS and N or S. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Reading compact DMS digits as a decimal number. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve original DMS form or explicitly verified conversion. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DECLATITUDE exclusive representation. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

21. DECLONGITUDE — meaning, non-equivalence and review

What the record says. DECLONGITUDE refers to longitude in decimal degrees, positive east. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Confusing east with west when a UI changes direction. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve signed decimal with datum. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DECLATITUDE and location bounds. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

22. LONGITUDE — meaning, non-equivalence and review

What the record says. LONGITUDE refers to alternative longitude in degrees minutes seconds plus E or W. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Treating longitude DMS as time or a street coordinate. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep source format and direction suffix. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DECLONGITUDE exclusive representation. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

23. COORDUNCERT — meaning, non-equivalence and review

What the record says. COORDUNCERT refers to coordinate uncertainty expressed in metres. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Converting uncertainty to the number of decimal places. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep uncertainty numeric value and unit as recorded. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are GEOREFMETH and coordinate precision. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

24. COORDDATUM — meaning, non-equivalence and review

What the record says. COORDDATUM refers to geodetic datum or reference system used for coordinates. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Automatically labeling old map-derived coordinates WGS84. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Retain stated datum; unknown remains unknown. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DECLATITUDE and DECLONGITUDE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

25. GEOREFMETH — meaning, non-equivalence and review

What the record says. GEOREFMETH refers to method used to determine collecting site coordinates. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Claiming GPS observation when location was estimated from a map. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Translate method as historical procedure with evidence. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COORDUNCERT and COLLSITE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

26. ELEVATION — meaning, non-equivalence and review

What the record says. ELEVATION refers to elevation at collecting site measured in metres above sea level. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Replacing source elevation with altitude from a new global DEM. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve source unit, sign and method if recorded. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLSITE and GEOREFMETH. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

27. COLLDATE — meaning, non-equivalence and review

What the record says. COLLDATE refers to date the material was originally collected. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Changing uncertain month to January or conflating with receipt. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Retain YYYYMMDD missing-part convention. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are ACQDATE and COLLMISSID. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

28. BREDCODE — meaning, non-equivalence and review

What the record says. BREDCODE refers to wIEWS institute code for entity that bred material. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Calling the breeding institute the donor in every case. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep role and code separate. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are BREDNAME, DONORCODE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

29. BREDNAME — meaning, non-equivalence and review

What the record says. BREDNAME refers to breeder organization or person name when no code is available. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Confusing named breeder with botanical species authority. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Translate human-readable description but preserve source form. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are BREDCODE, SPAUTHOR. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

30. SAMPSTAT — meaning, non-equivalence and review

What the record says. SAMPSTAT refers to controlled biological status such as wild, landrace or breeding material. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Rendering status as genetic quality or legal availability. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve numeric code and approved label. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLSRC, MLSSTAT. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

31. ANCEST — meaning, non-equivalence and review

What the record says. ANCEST refers to pedigree or description of ancestry, sometimes with symbolic crosses. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Normalizing breeding symbols as arithmetic or genealogy prose. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Protect cross notation and provide separate explanation. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are SAMPSTAT, BREDNAME. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

32. COLLSRC — meaning, non-equivalence and review

What the record says. COLLSRC refers to controlled collecting or acquisition source such as farm or wild habitat. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Replacing a farm store with farmer’s field or a shop with wild origin. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Respect source hierarchy and code. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are COLLSITE, SAMPSTAT. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

33. DONORCODE — meaning, non-equivalence and review

What the record says. DONORCODE refers to wIEWS code identifying donor institute. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Swapping donor and holding institution in target syntax. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve direction of transfer. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DONORNUMB, INSTCODE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

34. DONORNAME — meaning, non-equivalence and review

What the record says. DONORNAME refers to donor name if official institute code cannot be supplied. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Creating a fictitious code from a translated proper name. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep original name and authorized display translation. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DONORCODE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

35. DONORNUMB — meaning, non-equivalence and review

What the record says. DONORNUMB refers to accession designation assigned by donor before transfer. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Relabeling donor number as present holding accession number. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve donor-scoped identity and link. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are ACCENUMB, OTHERNUMB. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

36. OTHERNUMB — meaning, non-equivalence and review

What the record says. OTHERNUMB refers to paired institute and alternate identifier references. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Removing colon or semicolon delimiters that encode relationships. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve INSTCODE:identifier pairing. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are PUID and donor history. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

37. DUPLSITE — meaning, non-equivalence and review

What the record says. DUPLSITE refers to institution code of safety-duplicate holding locations. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Saying a duplicate is accessible for distribution or a separate accession. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve safety-duplication role, not public-request claim. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DUPLINSTNAME, STORAGE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

38. DUPLINSTNAME — meaning, non-equivalence and review

What the record says. DUPLINSTNAME refers to names of institutes maintaining safety duplicates. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Calling safety backup the main active holding genebank. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Describe redundancy accurately. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DUPLSITE, INSTCODE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

39. STORAGE — meaning, non-equivalence and review

What the record says. STORAGE refers to controlled scheme for seed, field, in vitro, cryogenic and DNA collections. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Equating storage mode with germination viability or availability. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve official numeric code and multiple-value syntax. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are AVAILABLE, CURATION. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

40. MLSSTAT — meaning, non-equivalence and review

What the record says. MLSSTAT refers to status under ITPGRFA Multilateral System. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Translating 0 as a ban on all transfer or 1 as unconditional access. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve status scope and any 99 remark. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are AVAILABLE and SMTA process. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

41. REMARKS — meaning, non-equivalence and review

What the record says. REMARKS refers to field-specific notes that elaborate exceptional values. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Deleting field-prefix or splitting semicolon syntax incorrectly. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Protect fieldname:remark structure, translate only authorized free text. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are Associated source descriptor. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

42. ACCEURL — meaning, non-equivalence and review

What the record says. ACCEURL refers to uRL to provider’s accession page. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Replacing record URL with translated marketing landing page. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep canonical link and evaluate redirect separately. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are PUID, UUID. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

43. AVAILABLE — meaning, non-equivalence and review

What the record says. AVAILABLE refers to genesys extension indicating material requestability when configured. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Assuming blank means 0 or interpreting 1 as guaranteed immediate shipment. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Preserve 0, 1 or unspecified distinction. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are HISTORIC, MLSSTAT. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

44. HISTORIC — meaning, non-equivalence and review

What the record says. HISTORIC refers to genesys flag for record no longer actively maintained. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Erasing archived record as irrelevant or treating historic as genetic age. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep historical record and state. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are CURATION and material requestability. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

45. CURATION — meaning, non-equivalence and review

What the record says. CURATION refers to genesys extension describing full, partial, archived or historical curation. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Interpreting fully curated as certified high-yield seed or legally cleared. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Use exact state semantics and provider precedence rules. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are HISTORIC, STORAGE, AVAILABLE. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

46. UUID — meaning, non-equivalence and review

What the record says. UUID refers to stable record identifier used by Genesys where applicable. In translation this is a typed claim, not a prompt for stylistic paraphrase. Its precise function matters because accession records can be aggregated, filtered, cited and compared by people who never read the prose surrounding them. Preserve the authoritative field name in the machine layer; a localized display label is a separate asset. Distinguish a verified fact in the source from a translator’s note about what an unfamiliar term usually means.

Characteristic failure. Moving UUID into PUID field or rewriting lowercase hexadecimal characters. A fluent target phrase can conceal this error because its words sound plausible, especially when the field resembles a familiar agricultural term. The reviewer should ask which object, institution, date, status or identifier the source has actually bound to this field. If the proposed wording would cause a different accession to be retrieved or a materially different research inference to be made, reject it even if it reads smoothly. An unverified expansion belongs in a query to the curator, never in the published target value.

Corrective procedure. Keep UUID in its own field and preserve original value. Keep a row-level before-and-after comparison and preserve the raw value next to any human-readable translation in the review worksheet. The related fields to check are DOI/PUID and accession composite key. These relationships are important: a contradiction may indicate source-data inconsistency rather than translator error. Do not repair the underlying database by assumption; document the conflict, request the authorized source decision and apply the decision only after review. For spreadsheet exports, check that quoted strings, separators, numeric formatting and empty cells survive the pipeline.

Failure modes: how apparently good translations damage genebank evidence

Failure 1: Accession number treated as an integer

Trigger: A local identifier 000712 is imported as 712. The immediate danger is not just a wrong word; it is a changed data relationship. The identity string may be different; spreadsheet type inference removed significant leading zeros. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Force text storage at import and compare raw code points with source values. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

A practical acceptance test is to describe the accession in the target language using only the restored evidence, then identify one claim the translation must not make. Write the forbidden inference in the QA ticket. This discipline is especially useful in multilingual teams, because a reviewer who does not know the original language can still check whether the target text overstates origin, legal status, biological identity or temporal precision.

Failure 2: DOI normalized into a title

Trigger: A DOI is translated into a human-friendly article title. The immediate danger is not just a wrong word; it is a changed data relationship. Persistent identity and DOI resolution are lost. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep the DOI untouched and use translated text only in a neighboring label. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 3: Different institute namespaces merged

Trigger: Two accession records share the suffix 481 but have different INSTCODE values. The immediate danger is not just a wrong word; it is a changed data relationship. A mistaken join creates a false duplicate or replaces a valid record. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Key by institutional namespace and original accession number. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 4: Collector mistaken for holder

Trigger: A collecting institute appears in the target as the current conserving institution. The immediate danger is not just a wrong word; it is a changed data relationship. Readers may contact the wrong repository or infer false custody. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Draw a holder-collector-donor graph and retest each sentence. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 5: Acquisition confused with collection

Trigger: The year an accession reached a genebank is called the year it was harvested from nature. The immediate danger is not just a wrong word; it is a changed data relationship. The record’s historical time line is corrupted. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep ACQDATE and COLLDATE distinct and preserve uncertainty. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 6: Country of origin replaced by storage country

Trigger: The target page describes the provider’s home as origin. The immediate danger is not just a wrong word; it is a changed data relationship. Provenance, adaptation analysis and geographic search become misleading. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Protect ORIGCTY as separate from holder location. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 7: Country code localized within the CSV

Trigger: ISO alpha-3 code is replaced with a translated country name. The immediate danger is not just a wrong word; it is a changed data relationship. Downstream filters no longer match the controlled schema. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Translate the display name only; leave code in data. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 8: Negative latitude lost

Trigger: A southern latitude becomes a northern latitude. The immediate danger is not just a wrong word; it is a changed data relationship. A site may be moved thousands of kilometres. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Validate sign, hemisphere, datum and plausibility jointly. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 9: DMS longitude read as decimal

Trigger: Compact 0762510W becomes 76.2510 degrees west without conversion. The immediate danger is not just a wrong word; it is a changed data relationship. Degrees-minutes-seconds data are misinterpreted. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Parse degrees, minutes and seconds as separate units before conversion. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 10: Precision manufactured from a map

Trigger: A site described only to nearest town receives five-decimal coordinates. The immediate danger is not just a wrong word; it is a changed data relationship. The translation fabricates observational precision. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Retain textual site and record coordinate uncertainty or leave numeric fields empty. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 11: Datum silently replaced

Trigger: An older map datum is labeled WGS84 without transformation evidence. The immediate danger is not just a wrong word; it is a changed data relationship. Coordinates appear precise but refer to the wrong reference frame. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Preserve named datum; escalate any transformation for geospatial review. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 12: Uncertainty converted to confidence

Trigger: COORDUNCERT 500 is rendered as 500 percent certainty. The immediate danger is not just a wrong word; it is a changed data relationship. A dimensional measurement is made nonsensical. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Name units and read it as uncertainty in metres. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 13: Altitude confused with plant height

Trigger: Elevation at collecting site is translated as the crop’s mature height. The immediate danger is not just a wrong word; it is a changed data relationship. Ecological metadata becomes a trait claim. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep environmental elevation apart from morphological measurements. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 14: Botanical authority mistaken for author credit

Trigger: SPAUTHOR abbreviation is used as bibliographic article author. The immediate danger is not just a wrong word; it is a changed data relationship. Taxonomic citation becomes misleading. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Preserve botanical-author role separately from documentation authorship. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 15: Subtaxon rank translated as commercial grade

Trigger: A taxonomic var. is interpreted as quality class. The immediate danger is not just a wrong word; it is a changed data relationship. The biological classification is changed. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep taxonomic abbreviations and explain them only where needed. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 16: Common name overwrites taxon

Trigger: An everyday crop name replaces GENUS/SPECIES. The immediate danger is not just a wrong word; it is a changed data relationship. Distinct taxa are collapsed under a broad name. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Retain scientific identity and localize CROPNAME separately. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 17: Species inferred from crop name

Trigger: A genus-only record receives a guessed species in translation. The immediate danger is not just a wrong word; it is a changed data relationship. Uncertain identification becomes invented certainty. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep sp. or blank according to source and request determination. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 18: Landrace equated with improved variety

Trigger: SAMPSTAT code 300 is described as advanced registered cultivar. The immediate danger is not just a wrong word; it is a changed data relationship. A biological-history category is reversed. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Use controlled MCPD codebook before translating the status label. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 19: Wild habitat confused with wild biological status

Trigger: COLLSRC wild habitat is copied into SAMPSTAT wild. The immediate danger is not just a wrong word; it is a changed data relationship. Collecting-site environment and germplasm biological class are distinct. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Review both fields independently. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 20: Pedigree symbols converted to arithmetic

Trigger: A breeding cross formula using slashes and asterisks is ‘simplified’ as fractions. The immediate danger is not just a wrong word; it is a changed data relationship. Reproductive ancestry stops being readable. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep source pedigree string and add a cautiously checked prose gloss. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 21: Donor accession becomes collecting number

Trigger: DONORNUMB appears in COLLNUMB after a field mapping shift. The immediate danger is not just a wrong word; it is a changed data relationship. Two identifier histories are conflated. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Map each to its named role and test with a known source record. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 22: Name aliases split into separate accessions

Trigger: Multiple ACCENAME variants become multiple rows. The immediate danger is not just a wrong word; it is a changed data relationship. Inventory counts are inflated. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep aliases within one accession record where source requires. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 23: Other identifiers lose institute prefix

Trigger: OTHERNUMB loses INSTCODE: prefix. The immediate danger is not just a wrong word; it is a changed data relationship. Cross-collection identity cannot be traced reliably. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Preserve paired namespace and identifier syntax. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 24: Safety duplicate called active supply

Trigger: DUPLSITE becomes a place where anyone can order samples. The immediate danger is not just a wrong word; it is a changed data relationship. The backup role is misrepresented. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Describe safety duplication without asserting public distribution. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 25: Storage mode mistaken for viability

Trigger: STORAGE 13 is translated as seeds definitely viable and ready. The immediate danger is not just a wrong word; it is a changed data relationship. Storage protocol does not itself certify present viability. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep conservation mode separate from testing or stock evidence. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 26: Cryopreservation mistaken for regular freezer

Trigger: STORAGE 40 is loosely described as ordinary freezer storage. The immediate danger is not just a wrong word; it is a changed data relationship. Conservation technology is made scientifically inaccurate. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Use approved cryopreservation terminology and no invented protocol. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 27: DNA collection equated with seed supply

Trigger: STORAGE 50 becomes ‘seeds available’. The immediate danger is not just a wrong word; it is a changed data relationship. Material type is changed. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Represent a DNA collection as DNA material, with any separate availability rules. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 28: Blank availability converted to no

Trigger: AVAILABLE missing becomes unavailable. The immediate danger is not just a wrong word; it is a changed data relationship. Unknown is incorrectly treated as a refusal. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep unspecified distinct from 0 and 1. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 29: AVAILABLE 1 treated as unconditional entitlement

Trigger: Material requestability is translated as guaranteed delivery. The immediate danger is not just a wrong word; it is a changed data relationship. Physical stock, agreements and provider opt-in may still matter. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Describe the defined flag without promising shipment. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 30: Historic record deleted from catalogue

Trigger: HISTORIC 1 is treated as an invalid record to drop. The immediate danger is not just a wrong word; it is a changed data relationship. Important provenance evidence disappears. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Retain the record with accurate historic state. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 31: CURATION FULL advertised as certified seed

Trigger: A full curation status becomes a commercial quality assurance claim. The immediate danger is not just a wrong word; it is a changed data relationship. The translation implies a certification not present in record. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Use Genesys curation definition without marketing inflation. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 32: MLS status turned into a complete legal clearance

Trigger: MLSSTAT 1 becomes ‘can be shipped anywhere for any purpose’. The immediate danger is not just a wrong word; it is a changed data relationship. Access or SMTA processes remain context dependent. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Preserve MLS membership and seek legal/access review for additional claims. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 33: Free-text remarks treated as schema commands

Trigger: REMARKS fieldname:note syntax is reinterpreted or stripped. The immediate danger is not just a wrong word; it is a changed data relationship. The note loses its reference to original field. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep field prefix and delimiters while translating only the approved note. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 34: Multi-value delimiters changed

Trigger: Semicolons become commas in fields where semicolons encode distinct values. The immediate danger is not just a wrong word; it is a changed data relationship. An importer sees one malformed value or too many columns. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Retain source syntax and display localized list separately. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 35: Incomplete date completed

Trigger: Year-only 19750000 becomes 1 January 1975. The immediate danger is not just a wrong word; it is a changed data relationship. Unknown month and day turn into false chronology. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Retain format representing incomplete precision. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 36: Personal names revealed unnecessarily

Trigger: A free-text translation expands initials into a guessed full collector name. The immediate danger is not just a wrong word; it is a changed data relationship. Privacy risk and invented identity increase. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Use source value and subject it to disclosure review. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 37: Sensitive locality exposed through translation

Trigger: Vague source narrative is expanded using map inference to exact coordinates. The immediate danger is not just a wrong word; it is a changed data relationship. At-risk populations may become easier to locate. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Avoid unnecessary precision; apply institution’s publication restrictions. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 38: Current taxonomy inferred from an algorithm

Trigger: Automatic taxon matching is presented as the curator’s verified original determination. The immediate danger is not just a wrong word; it is a changed data relationship. Source assertion and computational match become indistinguishable. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Show matching result as a separate derived field. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 39: API JSON boolean converted to string

Trigger: The flag true is written as the text ‘yes’ in a typed payload. The immediate danger is not just a wrong word; it is a changed data relationship. A parser or validation rule may reject the record. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Keep machine type, localize only view labels. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 40: Historic source spelling silently corrected

Trigger: A collecting locality name is modernized with no annotation. The immediate danger is not just a wrong word; it is a changed data relationship. The original archival evidence becomes hard to trace. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Retain original form and add qualified current name separately. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Failure 41: Catalog snapshot marketed as live stock

Trigger: A dated export is described as today’s available supply. The immediate danger is not just a wrong word; it is a changed data relationship. Readers are given unverified current expectations. A record can remain grammatically polished while that change propagates into search filters, specimen requests, scientific comparisons or conservation reports.

Repair and verification: Include snapshot date and prompt direct confirmation from holder. Locate the original field, recover the literal source value and identify whether the defect arose during translation, extraction, spreadsheet transformation or publication. The reviewer should record a counterexample: one fictitious record for which the incorrect wording would produce a wrong conclusion. Verify the correction both in the visible target-language text and in the exported structured fields. Do not alter real collection data without its owner’s authorization.

Worked diagnostics: sixty fictional accession translation cases

Every accession number, mission number, site note and record fragment in the exercises below is illustrative, not a real accession lookup or an authority to distribute material. The cases deliberately vary the fault and the evidence required for correction. They are designed for parallel source-target review, not for copying hypothetical identifiers into a production genebank database.

Worked diagnostic 01: A holding accession with leading zeros

Source fragment: ACCENUMB = ‘0000421’. Tempting but wrong rendering: The target catalogue displays accession 421. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The zeros are identity characters in a string, not decoration. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve the exact key and ask the holder whether zero padding is significant in its namespace. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 02: Two holders share the same identifier

Source fragment: Two INSTCODE namespaces each show ACCENUMB=’B17′. Tempting but wrong rendering: The translator merges both catalogue cards. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: A number alone cannot establish that they contain the same germplasm. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep a composite reference and cross-check donor or collecting identifiers. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 03: A GLIS DOI accompanies a local number

Source fragment: PUID contains a DOI, ACCENUMB a separate text ID. Tempting but wrong rendering: The translated page says DOI replaced the accession number. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Persistent external identification and local accession numbering coexist. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Show each field’s role and test resolver without changing the DOI. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 04: A collecting number links distant collections

Source fragment: COLLNUMB is the same on two holdings, ACCENUMB differs. Tempting but wrong rendering: A bilingual display labels both as identical accession records. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Shared collecting history is evidence of a possible relationship, not proof of current identical lots. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Retain original collection number and examine provenance paths. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 05: A mission identifier looks like a plant ID

Source fragment: COLLMISSID=’EXP-04′. Tempting but wrong rendering: The translated sentence calls the plant accession EXP-04. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The token labels a collection mission, not the genebank’s accession. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep mission identifier under its own field and link the samples separately. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 06: A collector changed institutional affiliation

Source fragment: COLLCODE names original institution; INSTCODE names current holder. Tempting but wrong rendering: A translation says the present institution collected the seeds. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Institutional custody and collecting activity happen at different times. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Reconstruct actor roles with source documents. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 07: Historic organization has no WIEWS code

Source fragment: COLLNAME is populated, COLLCODE empty. Tempting but wrong rendering: A translator fabricates a three-letter-number code. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Unknown registration is not an invitation to invent one. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Retain organization name and state official code unavailable. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 08: Species identification remains open

Source fragment: GENUS provided, SPECIES=’sp.’. Tempting but wrong rendering: The target page adds a species taken from common name. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The original record explicitly avoids a resolved species epithet. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve uncertainty and request curatorial confirmation. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 09: Common crop label differs by language

Source fragment: CROPNAME has a regional vernacular name. Tempting but wrong rendering: The target replaces the botanical genus with translated vernacular label. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Common-language categorization is not a taxonomic determination. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Localize crop label in a separate field. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 10: Botanical author resembles a human collector

Source fragment: SPAUTHOR has an abbreviated authority. Tempting but wrong rendering: The display lists it under collecting personnel. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Scientific name attribution and specimen collection are different roles. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Validate species authority in taxonomy context. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 11: Subspecies mistaken for cultivar

Source fragment: SUBTAXA uses subsp.. Tempting but wrong rendering: Translated output describes a registered agricultural cultivar. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Botanical rank alone asserts neither commercial release nor registration. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep botanical hierarchy. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 12: A taxonomic synonym is suggested by matcher

Source fragment: Source taxonomy and current reference database differ. Tempting but wrong rendering: The translated text silently replaces the source name. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: An inferred match is distinct from original determination. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Show source name and matched name as separate assertions. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 13: Multi-name accession contains semicolons

Source fragment: ACCENAME=’Name A;Name B’. Tempting but wrong rendering: The importer splits this into separate rows. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Multiple designations can belong to one accession. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Treat name list as one typed field. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 14: Accession name equals a person’s name

Source fragment: ACCENAME matches a person’s given name. Tempting but wrong rendering: The target changes its spelling to a familiar local form. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Proper designation may be identity-bearing. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep original designation and add translation note only if authorized. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 15: Year-only acquisition record

Source fragment: ACQDATE=’19880000′. Tempting but wrong rendering: The output displays 01/01/1988. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Month and day were never known. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Represent only the year to readers. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 16: Collecting date differs from acquisition date

Source fragment: COLLDATE 1986 and ACQDATE 1994. Tempting but wrong rendering: Target text says the sample was collected in 1994. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The eight-year interval may reflect transfer or custody. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep original dates and distinguish events. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 17: Origin distinct from donor location

Source fragment: ORIGCTY points to one country; DONORCODE belongs elsewhere. Tempting but wrong rendering: Translation treats donor country as origin. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The material can be transferred across borders without changing provenance. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Protect origin code and donor relationship. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 18: Historical country names in archive

Source fragment: Free-text COLLSITE refers to an earlier administrative geography. Tempting but wrong rendering: Translator substitutes current region as source fact. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Retrospective terminology needs temporal qualification. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Show historical source and optionally current interpretation. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 19: Locality described relative to road

Source fragment: COLLSITE says ‘near junction, approximate’. Tempting but wrong rendering: The target asserts exact GPS placement. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Narrative wayfinding is not survey precision. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve approximation. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 20: A protected collecting place

Source fragment: COLLSITE contains sensitive habitat detail. Tempting but wrong rendering: The translation adds exact public location. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Even valid data may require deliberate withholding. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Apply rights and harm screening before release. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 21: Latitude hemisphere reversal

Source fragment: DECLATITUDE = -7.4. Tempting but wrong rendering: Translated UI shows +7.4. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: A sign flip changes geographic hemisphere. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Check numeric sign and map plausibility. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 22: Longitude hemisphere reversal

Source fragment: DECLONGITUDE = -71.2. Tempting but wrong rendering: Target renders 71.2E. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: East/west is inverted. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Confirm signed decimal and label conventions. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 23: DMS compact longitude

Source fragment: LONGITUDE=’0762510W’. Tempting but wrong rendering: Machine interprets 76.2510°W. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: DMS is not a decimal digit string. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Parse as 76°25′10″W. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 24: Latitude and longitude swapped

Source fragment: Two coordinate fields are transposed. Tempting but wrong rendering: A map pin moves dramatically. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Each component has a different valid range and axis. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Verify field order with explicit known reference. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 25: Unknown coordinate datum

Source fragment: COORDDATUM blank. Tempting but wrong rendering: Target publication declares WGS84. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: A transformation cannot be assumed. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep datum unknown and flag for review. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 26: Map-derived georeference

Source fragment: GEOREFMETH indicates map estimate. Tempting but wrong rendering: Target says measured by GPS. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Method of inference is misreported. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve method label and uncertainty. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 27: Generalized coordinates

Source fragment: COORDUNCERT indicates broad spatial uncertainty. Tempting but wrong rendering: Target displays a precise decimal marker with no caveat. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Formatting precision is not evidence precision. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Expose uncertainty or withhold coordinates if sensitive. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 28: Site elevation below sea level

Source fragment: ELEVATION negative. Tempting but wrong rendering: Translator removes minus sign. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Negative elevation may be physically meaningful. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Retain sign and units. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 29: Collection from a market

Source fragment: COLLSRC coded as market. Tempting but wrong rendering: Target says gathered from wild forest. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Acquisition environment and biological category have diverged. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Use controlled COLLSRC mapping. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 30: Farm store versus field

Source fragment: COLLSRC for farm store. Tempting but wrong rendering: Target says sampled directly in a field. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The collection context matters for provenance. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep exact subtype. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 31: Natural wild versus semi-natural

Source fragment: SAMPSTAT has a specific wild subclass. Tempting but wrong rendering: Target generalizes it to a named cultivar. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Biological status hierarchy is lost. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve precise code level. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 32: Landrace record miscalled a breeder line

Source fragment: SAMPSTAT 300. Tempting but wrong rendering: Translation chooses breeding material. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Traditional cultivar and research line are different categories. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Use controlled labels. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 33: Cross notation within pedigree

Source fragment: ANCEST contains slash and asterisk pattern. Tempting but wrong rendering: Editor normalizes signs as punctuation. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: These characters may encode pedigree relationships. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Leave raw pedigree intact and annotate only with expertise. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 34: Unknown breeder identity

Source fragment: BREDNAME present, BREDCODE empty. Tempting but wrong rendering: Translator invents WIEWS code. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Organization identification requires registration evidence. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Use name-only route. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 35: Donor number repeated in local notes

Source fragment: DONORNUMB shown within REMARKS. Tempting but wrong rendering: Translator treats it as extra accession. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: One historic identifier can occur in multiple contexts. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Bind number to its donor. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 36: Other number pairs

Source fragment: OTHERNUMB=’AAA001:Q7;BBB001:R9′ illustrative. Tempting but wrong rendering: Target drops prefixes and only shows Q7 and R9. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The namespace is needed to understand each alternate reference. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve complete pair syntax. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 37: A backup deposit is not public supply

Source fragment: DUPLSITE names safety repository. Tempting but wrong rendering: Translated summary tells users to order directly from it. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Duplication is a risk-management relationship. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Refer requestability to the holding provider. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 38: Seed storage and cryo storage coexist

Source fragment: STORAGE=’13;40′. Tempting but wrong rendering: Target chooses only ‘seed bank’. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The record indicates more than one conservation mode. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve multi-value list. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 39: DNA material not reproductive seed

Source fragment: STORAGE=’50’. Tempting but wrong rendering: Target calls it a seed shipment. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: DNA holdings and living propagules are not equivalent. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Maintain material type. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 40: MLS participation not a shipping promise

Source fragment: MLSSTAT=’1′. Tempting but wrong rendering: Target says ‘available without restrictions’. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The code reports system participation, not complete transaction terms. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep access and provider process separate. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 41: MLS status explicitly unknown

Source fragment: MLSSTAT blank. Tempting but wrong rendering: Target uses ‘No’. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Unreported cannot be translated as negative. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Leave source unknown. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 42: REMARKS clarifies an other code

Source fragment: REMARKS=’COLLSRC:riverside’. Tempting but wrong rendering: Target removes field prefix. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The note loses the coded field it elaborates. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep prefix and translate approved free text. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 43: A URL points to old catalogue

Source fragment: ACCEURL resolves to redirect. Tempting but wrong rendering: Translator substitutes a new search homepage. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: A URL’s referent may have changed. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Validate canonical accession-level destination rather than hiding broken link. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 44: Unknown distribution state

Source fragment: AVAILABLE blank. Tempting but wrong rendering: Website says unavailable. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Unspecified is not identical to false. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Publish neutral status or seek update. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 45: Distribution marked yes

Source fragment: AVAILABLE=’1′. Tempting but wrong rendering: Translation promises immediate worldwide seed delivery. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Provider authorization and stock constraints still apply. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Use requestable interpretation, not unconditional entitlement. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 46: Historic accession retained as evidence

Source fragment: HISTORIC=’1′. Tempting but wrong rendering: Target removes record from historical catalogue. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Past accession records remain important. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve historic annotation. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 47: Historic and curation fields disagree

Source fragment: HISTORIC=’1′; CURATION=’FULL’. Tempting but wrong rendering: Target chooses historic without reviewing precedence. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Genesys documentation specifies field precedence. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Check current provider semantics and escalate mismatch. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 48: Partially curated accession

Source fragment: CURATION=’PARTIAL’. Tempting but wrong rendering: Translation calls material abandoned. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Partial maintenance is not equivalent to extinction. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Describe curation scope accurately. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 49: Archived material exists

Source fragment: CURATION=’ARCHIVED’. Tempting but wrong rendering: Target calls item destroyed. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Archived state may mean physical material remains. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Distinguish from HISTORICAL. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 50: Record-only historical state

Source fragment: CURATION=’HISTORICAL’. Tempting but wrong rendering: Translation says physical stock present. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: This category records the past without current physical material. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Preserve status and avoid request promise. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 51: UUID and DOI coexist

Source fragment: UUID and PUID separately populated. Tempting but wrong rendering: Target treats both as interchangeable resolver strings. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: They serve different identity mechanisms. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep both typed. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 52: Machine CSV header localized

Source fragment: MCPD header GENUS becomes translated label. Tempting but wrong rendering: Importer rejects column. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Data schema key must remain machine-readable. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Create separate display dictionary. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 53: Machine boolean localized

Source fragment: HISTORIC true converted to text ‘yes’. Tempting but wrong rendering: Typed export is no longer valid. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Preserve Boolean data type. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Translate only human interface. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 54: Comma decimal in CSV

Source fragment: Latitude -9.4 rendered ‘-9,4’ unquoted. Tempting but wrong rendering: CSV parser sees extra column. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Locale formatting conflicts with delimiter. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep invariant numeric output and localize view layer. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 55: Spreadsheet converts long identifiers

Source fragment: An accession ID beginning with zeros is numerically formatted. Tempting but wrong rendering: The public export omits required characters. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Input pipeline caused data corruption before translation. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Compare text-mode raw fields. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 56: One accession exported twice

Source fragment: Duplicate row created during bilingual joining. Tempting but wrong rendering: Reported collection total increases. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: A language join should not change entity count. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Use stable accession key and cardinality check. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 57: Trait dataset mixed into passport data

Source fragment: Plant height measurement appears in COLLSITE field. Tempting but wrong rendering: Environmental provenance and phenotypic evaluation are conflated. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Use a separate characterization/evaluation dataset. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: undefined. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 58: PDCI score used as quality grade

Source fragment: A 7/10 completeness index becomes ‘70% genetically pure’. Tempting but wrong rendering: Completeness does not imply biological superiority. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Define the metric’s actual information scope. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Worked diagnostic 59: An API filter described as full download

Source fragment: Bulk query and smaller filter endpoints are described identically. Tempting but wrong rendering: Readers may assume partial results are complete. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Use current documented pagination and request limits. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Worked diagnostic 60: Old export declared current

Source fragment: A 2024 source snapshot is translated in 2026. Tempting but wrong rendering: Readers think current availability was confirmed. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Show export timestamp and confirmation route. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Worked diagnostic 61: Collector name in REMARKS

Source fragment: Free-text note identifies an individual. Tempting but wrong rendering: Translator elaborates private biographical information. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: The disclosure risk increases beyond source. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Minimize and apply policy to publication. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 62: A third party supplied restricted records

Source fragment: A donor dataset has a confidentiality condition. Tempting but wrong rendering: Translation republishes everything because text was already shared internally. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Translation rights and public release rights are separate. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Review permissions before publication. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 63: Taxonomy matcher disputes original name

Source fragment: External backbone proposes different current name. Tempting but wrong rendering: Target replaces original field silently. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Automated suggestions must not erase source evidence. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Store derived match separately. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 64: An accession type may change over time

Source fragment: A regeneration yields a new lot. Tempting but wrong rendering: Target describes old viability result as present condition. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Temporal scope of a measurement matters. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Tie claims to evidence date. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 65: An accession URL has language parameter

Source fragment: Page route changes locale but not identity. Tempting but wrong rendering: Translator creates separate accession identity by language. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Locale view is not a new germplasm entity. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Keep canonical identifier and hreflang-like metadata. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Worked diagnostic 66: Fictitious identifier sneaks into production

Source fragment: A training row uses placeholder WIEWS-like code. Tempting but wrong rendering: Translator copies it into real export. This is an invented classroom record, deliberately stripped of any real accession entitlement. A translator should annotate which tokens are identity-bearing, which are free language, and what the receiving reader is likely to infer before deciding how to state the result.

Semantic mechanism: Illustrative values are not registered institute codes. The mistake matters because the source record and the translated version would answer different research questions. A catalogue user might search a different identifier, infer a different provenance chain, select the wrong environment, misunderstand an access condition or mistake a historical assertion for a current measurement. Correctness therefore requires more than a direct word substitution; it requires preserving the typed relationship that the field actually encodes.

Proposed repair: Use clearly segregated training fixtures and validation. Construct a two-column source-target check that includes the full field name, the unmodified machine value, an authorized reader-facing phrase and a brief reason for any explanatory gloss. Then test the statement against one adjacent descriptor that a non-specialist might confuse with it. Record a stop decision if the source evidence is insufficient. The released translation may be less specific than an invented completion, but it will remain scientifically defensible.

Transfer workshops: applying the same translation logic to other data tasks

Transfer workshop 1: Accession search results versus the source registry

A search-results snippet often shortens long record fields, omits missingness and promotes a common crop label. The editorial task is to restore the distinction between discovery and verification. Build a target-language search card with crop label, source accession key, official DOI if available and a link to the full record. Do not describe the search card as a certified data certificate. Ask reviewers whether it directs readers to the same underlying object and warns them before treating an excerpt as complete. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

A good transfer practice is to localize only the snippet labels and verify the exact link to the underlying record. Comparison of snippet and full record is more important than literary fluency. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 2: Research paper materials table

A paper may list accessions by codes that look like row numbers. Before translating the table legend, determine which codes identify holding accession, donor and sample within experiment. Retain immutable identifiers in every language and cite the source registry when possible. Where sample labels were locally assigned by a laboratory, avoid implying they are permanent globally recognized accession identities. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

A reviewer should be able to trace from table row to accession, from accession to institutional record, and from that record to any relevant material request rules. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 3: Multilingual germplasm request form

A request form combines descriptive text, fields entered by a user and authorization conditions. Translate instructions and error messages, not identifiers. Distinguish finding an accession, submitting a request, receiving approval and shipment. Do not turn a selection box labeled available into an unconditional promise that every request will be approved. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Test whether a reader in each language understands when an application has been submitted versus accepted, and which pieces of information are mandatory. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 4: Historical collecting expedition notebook

Old field notebooks may give only narrative locality, approximate year and a local specimen number. The priority is to preserve observation rather than retroactively fill modern structured fields. Keep diplomatic transcription separately from interpretation, annotate any modern equivalent as derived and avoid adding coordinates without documented method and uncertainty. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Test against the original page image when typed text could have mistaken a minus sign, direction or specimen digit. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 5: Geoportal point layer

Map displays are especially vulnerable to attractive but false precision. A site generalized for confidentiality must not appear as a precise point on a multilingual public map. Preserve datum, coordinate uncertainty and disclosure level, and choose a suitably coarse visual representation. Do not offset a point randomly while implying observational accuracy. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Check the map at the same zoom and in every locale; numeric coordinate tables and map pins must tell the same epistemic story. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 6: DNA sequence and barcode metadata

A biological sequence accession has its own identifier and provenance; it is not automatically identical to a genebank accession. Translate the explanatory link between the sample, voucher, sequence and germplasm record. Keep the sequence accession’s namespace and the genebank accession key distinct, even if both display the same taxon. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Test whether a reader can tell which material was sequenced, what identity was asserted and where the related physical resource is held. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 7: Trait dataset join

An evaluation dataset may record plant height, flowering time or disease response for accessions. The measurements may have environment, season and experimental design qualifiers. Preserve accession keys during joins and do not insert trait values into passport-origin fields. If a translated trait name changes scale or unit, the apparent scientific result may change. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Verify trait-unit consistency and source experiment identifiers, rather than only validating the translated column caption. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 8: Genebank conservation report

A summary may aggregate stored accessions, historic records, taxa or available material. Define the denominator in the target language and retain distinctions between physically maintained items and archival entries. Do not infer success of conservation from raw count alone; duplicate holdings and curation status matter. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

The best translation keeps denominators, reference date and inclusion criteria explicit. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 9: Safety duplication memorandum

A receiving repository may hold backup material without being the routine distributor. Translate the custodial role, obligations and transfer record without inventing public ordering routes. Protect donor, depositing institution, receiving repository and accession keys as separate entities. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

A test is whether a person could misroute a request or misread the backup as a new independent accession. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 10: Country-of-origin analytics

A database chart may group accessions by ISO alpha-3 ORIGCTY. Translate chart labels for the audience but preserve the actual country-code fields and period of collection. Origin can reflect collecting, breeding or selection, depending on material type; do not conflate country of genetic diversity with a simple label of ownership. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Check the coding, historical geography and reasons for missing values before drawing comparisons across countries. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 11: Crop taxonomy catalogue

A public catalogue may map common-language crop names to genera and species. When a name is used colloquially for several crops, do not choose a species by intuition. Show the original scientific determination, matched taxonomy if derived and vernacular aliases as different layers. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Ask whether an accession lookup using the scientific taxon yields the same record after localization. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 12: API integration documentation

The Genesys 2026 guides separate bulk passport access and trait-data retrieval. Translate parameter explanations for developers without altering endpoint paths, JSON field keys, Boolean types, pagination semantics or documented limits. API documentation can be informative even when the dataset it refers to has changed since publication. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Verify the target-language example against the current published API rather than copying an old screenshot as truth. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 13: Training a bilingual curator

A good exercise supplies a source excerpt, a target rendering and a deliberate trap. A learner must identify the field type, highlight invariant characters, name the adjacent descriptor being confused and propose a question to the source owner. Give credit for leaving an unknown field unresolved rather than inventing a polished answer. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Score factual identity preservation, terminology, uncertainty and auditability separately from style. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 14: Public outreach story about a landrace

A story may celebrate cultural heritage while drawing on accession records. Translate the narrative accessibly but avoid claiming that a single recorded accession represents all local cultivation traditions. A farmer’s traditional name can coexist with a formal accession ID; the two are not synonyms for every purpose. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Compare the narrative against the record and community-related publishing restrictions. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 15: Open-data release checklist

Before publication, inspect provenance rights, personally identifying details, endangered wild-relative sites and contractual embargoes. Translation review is not enough; release authority remains with the data holder. If generalizing location, reduce precision honestly and document uncertainty. Do not fabricate coordinates or use secret offsets presented as observations. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Test whether each public field is both linguistically accurate and authorized for disclosure. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 16: Machine translation quality audit

Machine output may preserve obvious capital letters yet still transform blank statuses into false, reformat dates or swap actor roles. Build targeted challenge cases for identities, taxonomic authority, dates, numerical signs and conditional access. A fluent score is not evidence of safe data equivalence. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Use analytic error categories and human evaluator judgment suited to the document’s purpose, consistent with ISO 5060:2024’s general evaluation approach. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 17: Multilingual source correction workflow

When the source curator corrects an accession record, the translation team should receive a change log identifying the field, record, old value, new value, effective timestamp and authority for correction. Only retranslate affected narrative fields; ensure immutable codes remain unchanged where the correction did not affect them. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

A versioned correction is better than quietly overwriting historical translation without trace. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

Transfer workshop 18: Educational glossary construction

A glossary should distinguish accession, sample, germplasm, genotype, landrace, cultivar, biological status, collecting source, donor, holder, duplicate, DOI and curation state. The same English word can have misleading equivalents in another language, so define concepts first, then test terms in complete sentences. The transferable principle is to keep the identity of what is described separate from the rhetoric used to describe it. A translator can simplify a sentence, but should not simplify away provenance, measured uncertainty, institutional role, material state or authority. Draft a compact evidence ledger with the source claim, available supporting field, target phrase and unresolved issue. This ledger makes disagreements inspectable rather than personal.

Protect field-level semantics while teaching how explanations change for an expert, student and general reader. For practice, create a fictional source row with one correct and one incorrect description. Give two reviewers different language versions and ask them to choose which one would lead to the same concrete action: identify, request, cite, preserve or defer. Score the explanation against the source evidence, not against fluency alone. If the readers agree on wording but infer different actions, the translation still needs repair.

A practical multilingual data contract and QA worksheet

A strong delivery contract has three parts. First, a machine-data contract: immutable MCPD field names, canonical identifiers, accepted value formats, controlled vocabularies, null handling, encoding and row-level key integrity. Second, a language contract: approved concept definitions, permitted display aliases, capitalization rules, scientific names and documentation of ambiguous local-language terms. Third, an evidence contract: the version and origin of each dataset, allowed use, precise dates, authorized corrections and scope of publication. These contracts work together because neither semantic accuracy nor machine validation alone is sufficient.

One workable review worksheet contains: source system; export timestamp; source accession composite key; field code; original value; field type; proposed target caption; proposed translated narrative; locked value; translator comment; second reviewer finding; curator decision; rights/sensitivity disposition; issue severity; resolved date; and release version. A bilingual reviewer should be able to understand the exact original claim without losing source wording. A data engineer should be able to reproduce the serialized output and determine whether null values or numerical signs changed.

Priority should be assigned by harm, not merely frequency. A wrong display synonym may be a moderate usability defect; a flipped coordinate, altered accession key, mistaken legal permission or exposed sensitive locality can be critical. Distinguish error severity from an overall linguistic score. The purpose of sampling and analytic error classification under ISO 5060:2024 is to support appropriate human evaluation, not to declare that a universal error threshold automatically guarantees biological data safety. Decide acceptance rules with the intended use and responsible institution.

Terminology clinic: twenty-five distinctions translators must be able to explain

1. Accession versus sample

An accession is a managed collection entity; a sample is material taken from or representing something, and may have its own handling history. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

2. Accession versus seed lot

A seed lot is a physical batch with particular production or handling history; a collection accession can be represented by different lots across time. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

3. Genebank versus seed company

A genebank emphasizes conservation and documentation; a commercial seed supplier operates under different objectives and evidence systems. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

4. Genebank accession versus certified seed

Passport identity does not by itself establish varietal certification, germination percentage, disease freedom or shipment eligibility. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

5. Passport data versus trait data

Passport data identify and contextualize holdings; trait observations record characteristics under conditions and methods. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

6. PUID versus local number

Persistent global reference and local collection identifier can coexist without being equivalent. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

7. DOI versus website address

A DOI identifies a resource persistently through a resolution framework; a record-page URL is a location that may change. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

8. WIEWS institute code versus institute name

The code supports unambiguous registry identification, while organization names are translatable and may change. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

9. Holding versus collecting institute

The present keeper and original collector are different functional roles, sometimes occupied by one organization. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

10. Collecting number versus mission ID

A specimen- or event-level collector designation should not be conflated with the organizing expedition’s identifier. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

11. Donor number versus current accession number

Different institutions may apply different cataloguing identifiers to related material. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

12. Country of origin versus storage country

Provenance of selection or collecting is not equivalent to the home of the present repository. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

13. Collecting place versus coordinates

A narrative location and numeric georeference may carry different precision, evidence and disclosure rules. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

14. Coordinate precision versus uncertainty

Decimal digits alone do not quantify how well a locality is known. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

15. Geodetic datum versus location description

A datum describes spatial reference; a place description describes locality in words. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

16. Elevation versus height of plant

One describes collecting-site vertical position; the other is a biological trait. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

17. Collecting date versus acquisition date

Original field activity and receipt into a collection occur as distinct events. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

18. Scientific name versus common name

The former is a taxonomic assertion with formal structure; the latter is audience-language vocabulary. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

19. Species authority versus named collector

Taxonomic name authorship and collecting activity must never be merged. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

20. Landrace versus advanced cultivar

A traditional cultivated population and an improved cultivar encode different histories. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

21. Biological status versus collecting source

What the material is and where it was obtained are separate descriptors. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

22. Storage versus viability

Conservation method is not equivalent to a current germination test or evidence of living material. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

23. Safety duplicate versus distributor

Backup custody does not imply rights or capability to fulfil routine requests. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

24. MLS inclusion versus automatic release

A system-level legal classification does not itself describe every transaction condition. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

25. Available versus confirmed dispatch

A requestable flag indicates an aspect of service state, not that shipment has been approved or completed. A clear target-language formulation should describe which question each concept answers and which inference it does not authorize. Use a plain-language definition for general readers, retain the technical label for specialist readers and never change its associated machine field. Where language offers one term for two different concepts, add a short qualifier rather than forcing a misleading one-to-one equivalence. The test is whether both a scientist and a catalogue user can distinguish the objects or states without access to the translator’s private notes.

Quality assurance diagnostics: what to measure

QA measure 1: Identity preservation

Count distinct composite keys before and after release; investigate every lost, duplicated or edited identifier. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 2: Referential integrity

Test DOI and accession-page links and verify they refer to the intended records, not merely to a functioning homepage. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 3: Codebook compliance

Inspect every controlled code against the actual schema version, especially SAMPSTAT, COLLSRC, STORAGE, MLSSTAT and service extensions. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 4: Precision conservation

Compare complete and incomplete dates, signed coordinates, spatial uncertainty and units for changes not supported by the source. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 5: Taxonomic fidelity

Check source scientific names and author strings against the original assertions, with derived synonym matches documented separately. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 6: Role fidelity

Ask which institution collected, bred, donated, holds or safely duplicates the accession; confirm no role is transferred by grammar. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 7: Material-state clarity

Check that historic, archived, partial, fully curated and distribution states remain distinct. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 8: Disclosure rights

Identify personal, sensitive geographic and embargoed information and document authorization for public output. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 9: Technical serialization

Round-trip representative CSV and JSON examples to detect type conversion, encoding and delimiter damage. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

QA measure 10: Audience usability

Give a genuine reader a task such as finding the correct record, understanding an unknown status or identifying where to request clarification. The measurement should be reproducible from a fixed test set, not an impression of fluency. Record the sample frame, the critical cases deliberately included, the nature of each deviation, what was corrected and which values remained unresolved. Report structural error counts separately from wording defects. When risk is high, expand review to the affected records rather than extrapolating from a small random sample.

Four-week practice plan: build competence rather than memorize labels

Week 1 — identity and terminology

Work through the holding institute, accession identifier, collecting identifier, DOI and scientific-name fields. Translate ten fictional record cards into a target language while locking machine values. For each card, draw the relationships among holder, collector and donor. Finish with a paired-language retrieval test and a glossary in which every term has a definition, an opposite confusion and a practical example.

Document completed exercises with a row-level evidence trail. An evaluator should be able to reproduce the correction and explain why a superficially smoother alternative was rejected. Continue the work only after the identity-bearing fields and source permissions are secure; nuanced stylistic improvement belongs after scientific integrity.

Week 2 — evidence, time and place

Practice incomplete dates, origin, collecting-site descriptions, decimal coordinates, DMS, datum, uncertainty and sensitive-location review. Use deliberately defective fixture records: a flipped hemisphere, an assumed January date, an approximate location presented as GPS and a hidden map-derived estimate. Fix them without inventing details. Review what should not be published as well as what should be translated.

Week 3 — status and access

Study biological status, collecting source, storage, safety duplication, MLS membership, availability, historic records and curation states. For every field create a positive, negative and unknown case. Write multilingual catalogue prose that distinguishes a discoverable accession from a requestable and shipped one. Escalate any apparent legal interpretation to the authorized office rather than translating a code as blanket permission.

Week 4 — production pipeline

Build a miniature fictional dataset with stable row keys, a versioned glossary, a field crosswalk, a source snapshot and two target-language displays. Validate CSV/JSON encoding, run semantic and structural QA, and produce a release report with unresolved issues. Simulate a subsequent source correction and verify that only affected translation assets are revised. The outcome is a reproducible procedure, not just a one-time document.

Frequently asked questions

Question: Can I translate MCPD field codes such as ACCENUMB?

Answer: Not in a machine-readable dataset. You may translate the user-facing caption, but the accepted schema key remains fixed.

Question: What is the minimum information for a Genesys accession record?

Answer: Genesys documentation identifies INSTCODE, ACCENUMB and GENUS as mandatory core identifiers. Other fields improve findability and context.

Question: Does every accession have a DOI?

Answer: No. DOIs are available as persistent identifiers through GLIS, but local accession identifiers continue to matter.

Question: Is a DOI the same as an accession number?

Answer: No. A persistent externally resolvable identifier and a locally assigned accession number can identify the related resource in different systems.

Question: Can I infer a missing species from a translated common name?

Answer: No. A genus-only or uncertain species identification must not become a confident taxonomic claim through translation.

Question: Should I translate botanical Latin names?

Answer: Preserve the scientific taxon and authority according to source and nomenclatural practice. Translate explanatory common names separately.

Question: Do I need to translate the country code?

Answer: Preserve the three-letter ISO country code in data. A target-language name can be shown as a display label.

Question: How should an unknown date be displayed?

Answer: Retain the level of knowledge present in the source; an unknown month and day cannot be turned into a specific calendar date.

Question: Does four decimal places mean a point is accurate to ten metres?

Answer: It can imply fine display precision, but location accuracy depends on measurement and source uncertainty. Preserve COORDUNCERT and the georeferencing method.

Question: Can we make exact collection coordinates public?

Answer: Only after authorized disclosure review. Some locations may need honest generalization or omission with uncertainty handled correctly.

Question: Is COLLSRC the same thing as SAMPSTAT?

Answer: No. One describes acquisition/collecting environment; the other describes biological or breeding status.

Question: Does seed storage indicate germination success?

Answer: No. Storage classification and viability test outcomes are different claims.

Question: What does AVAILABLE blank mean?

Answer: In the Genesys extension it is unspecified, not necessarily unavailable.

Question: Does AVAILABLE 1 guarantee shipment?

Answer: No. Requests may still depend on stock, provider participation, applicable agreements and further approval.

Question: What is the difference between HISTORIC and CURATION?

Answer: They are Genesys-specific state fields with documented semantics and precedence. Review that documentation before deciding how to summarize a record.

Question: Does MLSSTAT 1 mean no conditions apply?

Answer: No. It marks participation in the treaty’s Multilateral System; it does not replace transaction-specific guidance.

Question: Can a translated accession name replace the original?

Answer: Preserve the original designation as the identity-bearing name. Localized explanation can be added without overwriting it.

Question: Why preserve old records that are no longer active?

Answer: They can be needed to understand provenance, earlier publications and relationships to other collections.

Question: What if a holding institute and collecting institute are the same?

Answer: They may be represented by the same code, but the roles remain conceptually different and should stay distinguishable.

Question: What if the source record contradicts itself?

Answer: Escalate to the data owner, preserve the original, record the issue and do not silently harmonize factual differences.

Question: Can machine translation handle these records?

Answer: It can assist with selected human-readable fields, but locked tokens, typed values and specialist review are essential.

Question: Should we create translated CSV column headers?

Answer: Use separate target-language display mappings; maintain canonical field headers for import and exchange.

Question: Why do some records not reveal exact collection sites?

Answer: Data protection, conservation concerns, agreements and sensitive heritage information may require limits on disclosure.

Question: What is PDCI?

Answer: The Passport Data Completeness Index is a diagnostic of record completeness, not a measure of genetic purity or the scientific correctness of every claim.

Question: Where can I check current authoritative descriptors?

Answer: Use the FAO/Bioversity MCPD reference and Genesys provider documentation, not a translated glossary without provenance.

Question: How should a translation be evaluated?

Answer: Use a risk-based human evaluation of semantic and structural errors; ISO 5060:2024 offers general translation output evaluation guidance.

Question: How does this guide connect to vocabulary learning?

Answer: Terminology learning helps distinguish close concepts, but the protected vocabulary curriculum need not be rewritten to explain accession data.

Question: What is the most important final check?

Answer: Confirm that a reader using the target text would identify the same accession and infer no stronger claim than the source supports.

Evidence, standards and source dates

FAO/Bioversity MCPD 2.1 descriptor reference via Genesys documentation — source field meanings, codes and format conventions; original MCPD v2.1 issued December 2015.
Genesys: preparing MCPD-compliant data — data structure, required headers, coordinate convention, validation and common formatting errors.
Genesys: new passport and trait API guides (30 July 2026) — distinguishes bulk passport records from trait datasets and highlights field mapping.
FAO International Treaty: GLIS digital object identifiers — public reference on persistent identifiers for PGRFA.
FAO GLIS FAQ — DOI relationships with existing accession identifiers and transfers.
Genesys: what is passport data? — disclosure, collecting-site sensitivity and provenance caution.
Genesys: curation and validation — terminology, geographic validation and Passport Data Completeness Index.
Genesys: minimum data requirements — mandatory and recommended accession fields.
ISO 5060:2024 — general guidance for human evaluation of translation output, published February 2024.

These references were consulted as of 31 October 2026. Online software documentation and datasets can change. Confirm the current version of operational codebooks before using any example in production. This article paraphrases source terminology for educational use and does not reproduce proprietary standards verbatim. Fictional diagnostic cases are marked as exercises, not as research observations.

Conclusion: transfer meaning without inventing evidence

The master skill in plant genetic resource translation is to know which elements are words and which are evidence-bearing data. An accession identifier, an institute code, a botanical attribution, a collecting uncertainty, an availability flag or a status enum cannot be made more fluent by changing its claim. A responsible translator protects these facts, makes the surrounding explanation understandable, and leaves a readable audit trail. The result allows scientific users to retrieve the same resource across languages while distinguishing source facts from later interpretations.

The repeatable method is therefore to identify the accession; verify the schema and provenance; lock typed values; translate supported narrative fields; preserve uncertainty; check privacy and rights; test relationships, retrieval and serialization; and publish an accountable version. When a step fails, stop at the unsupported claim instead of inventing a completion. This is how multilingual documentation becomes trustworthy research infrastructure rather than merely good-sounding prose.

Next routes: translation architecture and protected learning hubs

Return to the parent Master Art of Translation | The Complete System for Moving Meaning Between Languages for the general translation method and neighboring specialist systems.
Use the protected Vocabulary Learning Hub for transferable vocabulary-learning methods, including distinctions, retrieval and accurate word choice.
Visit the protected How English Works | Translation and Meaning Across Languages for how context, grammar and explanation shape meaning without taking ownership of this article’s scientific content.

Series child: EDKSG-TRANS-MASTER-WORLD-830. Publication scope: new standalone world-facing article; no new claims about current availability of individual real accessions; no existing owner pages or navigation changed.