Skip to content
Language:DeutschEnglish Last updated: 2026-11-08

Export matrix

An overview of which Anton data is output via which export format — and which can only be reconstructed from the complete SQL dump.

What each artefact is good for is set out in Digital archive: overview. The short version: the native round trip and the SQL dump are backups, everything else is a publication or exchange view.

Status

This matrix reflects the state of 8 November 2026. It is maintained together with the exporters.

Available formats

Format Standard Scope Entry point
EAD 2002 apeEAD (ead.xsd + apeEAD.xsd) Object tree from fonds level Admin → Export
EAD3 ead3.rng Object tree from the fonds Admin → Export
EAD holding guide apeEAD type=holdings_guide Holdings overview Admin → Export
TEI – authority lists TEI-P5 standOff Actors / places / keywords (all) /api/tei/{actors,places,keywords}
TEI – per object TEI-P5 one record API with ?format=tei
SKOS – keywords SKOS (Turtle, RDF/XML, N-Triples) all keywords, a keyword with everything below it, or a concept scheme (with notation and the thesaurus URI): labels, broader and narrower terms, «see also», authority matches⁵ keyword list and keyword → «Download SKOS», Admin → Concept schemes, /skos/keywords.{ttl,rdf,nt}, /skos/schemes/{id}.{ttl,rdf,nt}; only with keyword_relations
RDF «A+» CIDOC CRM + RiC-O (dual) Object tree, all entities Admin → Export, anton:export-rdf
RDF RiC pure RiC-O 1.1 (JSON-LD) Object tree Admin → Export
RDF Memobase RiC-O subset (JSON-LD) Object tree (lossy) Admin → Export
RDF «A+» bundle CIDOC CRM + RiC-O + media files as a ZIP Object tree, viewable offline without Anton; with --include-protected --include-originals a migration package (see below) Admin → Export, anton:export-rdf --profile=a-plus-bundle
Native round trip anton-import-format v0.10 (JSON) + master media as a ZIP Object tree, losslessly re-importable anton:export-native / anton:import-native
Dublin Core OAI-DC per object (embedded only) Building block in DIP / OCFL
DIP BagIt Package: media + metadata (subtree) «DIP» button on the record
E-ARK DIP E-ARK DIP 2.2.0 (CSIP) Package of a record and its subtree: access = public view with access copies, full = everything with masters and originals «E-ARK DIP» button on the record (access), anton:export-dip --format=eark [--full]
OCFL Oxford Common File Layout Package: media + metadata (object/subtree) Button on the record
Excel (full) XLSX current result list, all fields Result list → Export
Excel (update table) XLSX Result list, only writable fields, re-importable Result list → Export → Update
Word / PDF DOCX / PDF Finding aid per object Button on the record
Paper HTML print view Result list (max. 1000) Result list
Preservation strategy preservation-strategy schema 1 (JSON) one decision of the preservation mandate, bytes as confirmed, with SHA-256 Admin → Digital preservation → Preservation mandate
SQL dump mysqldump (gzip) whole tenant database — without sessions, job queues and caches, API and upload tokens empty Admin → Export

Entity matrix

Key: ● full · ◐ partial / embedded · ○ not included

Entity EAD2002 EAD3 TEI RDF A+ RiC Memobase DC DIP/OCFL Excel full Excel update SQL dump
Unit of description ● ● ◐ ● ● ● ● ● ● ◐ ●
Hierarchy ● ● ○ ● ● ● ◐ ● ◐ ○ ●
Media (metadata) ◐ ◐ ◐ ◐ ◐ ◐ ◐ ● ○ ○ ●
Media (files) ○ ○ ○ ◐¹ ○ ○ ○ ● ○ ○ ○
Full text / OCR ○ ○ ○ ●¹ ○ ○ ○ ○ ○ ○ ●
Actors (embedded) ● ● ◐ ● ● ○ ◐ ◐ ◐ ◐ ●
Actors (authority file) ○ ○ ● ◐ ◐ ○ ○ ○ ○ ○ ●
Places (embedded) ◐ ◐ ◐ ● ● ○ ◐ ◐ ◐ ◐ ●
Places (authority file) ○ ○ ● ◐ ◐ ○ ○ ○ ○ ○ ●
Keywords ● ● ● ● ● ○ ◐ ◐ ◐ ◐ ●
Keyword relations⁵ ○ ○ ○ ○ ◐ ○ ○ ○ ○ ○ ●
Events (graph) ◐ ◐ ◐ ● ● ○ ○ ○ ◐ ○ ●
Text fields (ISAD fields) ● ● ◐ ◐ ◐ ◐ ◐ ◐ ● ● ●
Termselect values ◐ ○ ○ ● ○ ○ ○ ◐ ○ ○ ●
Pages / editions ○ ○ ○ ◐ ○ ○ ○ ○ ○ ○ ●
Languages / display date ● ● ○ ● ● ○ ○ ○ ◐ ◐ ●
Location (physical) ◐² ◐² ○ ● ○ ○ ○ ○ ● ● ●
Protection period (structured) ◐² ◐² ○ ● ○ ○ ○ ○ ● ● ●
Comments (internal) ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ●³
User accounts ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ●
Settings / forms ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ●
AI cataloguing data ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ●
File provenance (PRONOM/NARA) ○ ○ ○ ○ ○ ○ ○ ◐⁴ ○ ○ ●

² EAD 2002 and EAD3 write the closure period as a sentence in <accessrestrict>, with the year from which the material is open, next to the text field «Conditions of access». apeEAD allows no element of its own that could be evaluated. The text field «Location (detail)» goes out as <physloc>, the location reference does not.

³ Comments are internal working notes and deliberately appear in no publication format — not even in the native round trip. They are held exclusively in the SQL dump. This is not a state of implementation but a decision: they are kept in a separate table that no exporter knows about, so that they cannot accidentally end up in a publication either. See comments.

¹ Only in the A+ bundle: the pure A+ export delivers bundle-relative media references, image dimensions, AV duration and OCR full text in the graph; the ZIP additionally places the derivatives (thumb/web) alongside — in the standard case not the masters. With --include-originals the master files are added, see migration export.

⁵ Broader and narrower terms, «see also» (anton#706, #707) go out in full only in the SKOS export, between keywords: every broader term, not only the primary one, skos:related for «see also», skos:exactMatch for linked authority files (Wikidata, GND, LCNAF, VIAF, BnF, IdRef, GeoNames, SUDOC; not encyclopaedias, Wikipedia or images). The concept URI is the lasting address /id/{uuid}. The RiC export writes skos:broader, skos:narrower and skos:related between the concepts it holds and brings along the broader terms of the keywords in use. The native round-trip (format v0.10 on) carries every relation between keywords, actors, places and records without loss; a subtree carries the relations of what it holds, with the broader terms above. Relations to pages are only in the SQL dump.

⁴ PREMIS 3 per record with media (anton#661): formats (PRONOM), checksums, kept originals and every event on the media with tool and version. The NARA risk assessment is not included. Applies to the native round trip too, not to the ZH variant of the DIP.

Well covered

Units of description are broadly supported. EAD2002 is the most complete format: title also in the other languages, reference code, dating with its comment, level, extent, languages and script, closure period and the ISAD(G) text fields that the record's form shows to the public. Access points are written by type (persname, corpname, famname, subject, geogname) with the GND number where one is linked. Digital objects go out as the web view and thumbnail, never as the master. Every file is checked against ead.xsd and apeEAD.xsd as it is written, also in the nightly run; a finding is in the log and in the output of anton:export. Multilingual text fields are not possible: apeEAD allows each text field only once and has no language attribute on it. The version in the archive's main language goes out. RDF A+ types the same thing twice (CIDOC CRM and RiC-O), including multilingual titles.

Authority data has exactly one structured, standalone export: the TEI authority lists with all name forms, life dates, coordinates, description, sources and external links. Two restrictions: only free authority records not bound to objects are output, and actors or places without a type are skipped. Otherwise authority data appears only embedded in object exports.

Dates of actors go out as precisely as they were recorded: a year as a year, a month as a month. TEI carries a circa date as cert="medium", CIDOC CRM through the inner bounds (P81), RiC-O in the readable expressedDate. The native round trip and the actors spreadsheet keep it. CIDOC CRM gives the formation and dissolution of corporate bodies (from Anton v0.103.0).

Only in the SQL dump

This data survives no format export:

  • User accounts and authentication — which does not belong in an archival export anyway
  • Settings, form definitions, locations, orders, notifications — for locations also the arrangement of the magazine: levels, the place above and capacity (anton#832)
  • NARA risk assessment of the files — the rest of the file provenance (formats, checksums, events) travels as PREMIS in the DIP, OCFL and native round trip since anton#661
  • AI cataloguing data — profiles, consumption, budgets, audit samples
  • Editions and correspondences
  • The full-text index
  • Relations to pages — the native round-trip carries the other relations (anton#706)
  • The event graph as a relation — events appear only flattened (Excel) or implicitly (EAD, RDF)

Migration export: the bundle as a way out

The matrix above describes the standard case. The RDF export knows two switches that lift it out of the publication corner:

Switch Effect
--include-protected bypasses the data protection filter — withheld records, internal-only media and internal text fields come along. Takes effect with a-plus, ric and a-plus-bundle, not with memobase
--include-originals a-plus-bundle only: packs the original masters into the ZIP, not just web/thumb
php artisan anton:export-rdf --env=<slug> --root=<id> \
    --profile=a-plus-bundle --include-protected --include-originals

The result is a standards-compliant package of CIDOC CRM/RiC-O graph, blocked data and original files — suitable for transferring a fonds into another system. For the way back into Anton there is no ready-made tool — the native round trip remains responsible for that. And it contains personal data: do not host it publicly.

Who can trigger which artefact

Artefact Self-service in the interface
SQL dump yes — Admin → Export
EAD, TEI, RDF (standard), A+ bundle (standard), DIP, E-ARK DIP (access), OCFL, Excel, Word/PDF yes
Preservation strategy (JSON) yes — all admins, Admin → Digital preservation → Preservation mandate
E-ARK DIP full (--full) no — CLI only
A+ bundle with --include-protected / --include-originals no — the UI and the API do not pass the switches through, CLI only
Native round trip (anton:export-native / anton:import-native) no — a pure CLI command

With Anton as a Service, k & r produces the CLI artefacts on request; on premises, the commands are run by the operating institution.

Format restrictions in detail

EAD3 contains the same as EAD2002, written in EAD3 and checked against ead3.rng as it is written: extent and dating structured, names with @identifier and @source of the authority record, titles in the other languages with @lang. Text fields go out in the main language, as in EAD2002. The validation page checks EAD3 files too.

The EAD holding guide delivers a flat holdings overview without text fields per node.

Memobase RDF is deliberately lossy: no actors, places, keywords or events — only institution, objects, instantiations and around eight text field types.

TEI per object knows no overall export, no hierarchy and hardly any media.

E-ARK DIP (anton#697) comes in two scopes. Access is a publication view: what a guest may see, whoever exports — locked records and drafts are left out with their subtree, records under a protection period keep their description but not their media, private media are left out. The media are access copies (web) in representation access, described in EAD3 or EAD 2002 (setting eark_dip_ead), optionally also RiC-O and CIDOC CRM as Turtle (eark_dip_rdf). PREMIS names each representation's function and each file's source, without events and without persons. Full (--full, CLI only) is a hand-over like DIP and OCFL: every record, masters in master, kept originals in original, plus the native format (anton-native), PREMIS per record and the Word finding aid. Every package is validated before it is delivered; an invalid one is not delivered. See DIP download.

DIP, OCFL and native round trip carry a PREMIS 3 file per record with media since anton#661 (DIP: data/meta/<record>.premis.xml, OCFL: metadata/premis.xml, native: premis/<record>.xml). It lists the files and kept originals with MD5/SHA-512, size and PRONOM format, every event on the media (virus check, integrity and format checks, format identification, format rules, transfer to the cloud, ingest with the person who acted) and the tools involved with their version. A format rule at ingest is a normalization and names as implementer whoever added the medium; media:migrate is a migration and names the account that started the run, as <account>@cli. Every other event names, since version 0.117.0, whoever set it off (person, <account>@cli:<command> or job:<class>); older events name only the tool (anton#794). A file a format rule replaced without keeping it is an object of its own without a file, with the checksums, size and format the rule recorded before converting; the medium refers to it as its source (anton#751). Since version 0.118.0 each migration has a rights statement (rightsStatement, act «migrate») naming its ground: preservation mandate, confirmation by mail or agreement made outside Anton, linked to the migrated files, with the archive as grantor and Kränzle & Ritter as grantee (anton#807). Since version 0.119.0 a statement resting on the preservation mandate also names the version and SHA-256 of the strategy document in force at the time of the migration (strategy <version> sha256 <checksum>); for decisions without a document this is left out. There is no METS. The ZH variant of the DIP contains only media files, no metadata and no PREMIS.

The two Excel exports pursue opposite purposes. The full export is an evaluation view: it takes along everything that can be represented flatly and cannot be loaded back in. The update table is the opposite — it deliberately contains only what a data update is allowed to write, and leaves out hierarchy (parent) and events, because these would be blocked on re-import. The location is included: it is written on update, and rearranging the stacks is one of the most frequent reasons for a bulk update. The update table carries it as an ID, the full export since version 0.123.0 as its path in the magazine («Room 1 › Shelf 3 › Compartment 4»), a flat location still by its name (anton#832). Taking the narrower file therefore does not lose information by accident; it deliberately limits the scope of effect, and the download dialogue allows this to be narrowed further.

The update table of a multilingual archive carries the title in one column per content language (title_de, title_fr …) instead of a single one. This makes the round trip lossless in the multilingual case too: with one column, the language of the run would have to decide on re-import where the value belongs, and a French title would end up in the German field. Monolingual archives keep the bare titel column. The full export continues to output the title monolingually — it is a reading view, not a way back.

Neither is a backup: even the full Excel export carries neither media nor authority files nor forms. For a way back, the native round trip or the SQL dump serve (see below).

RDF A+ and the native round trip are two products

The A+ graph is a discovery view, not a backup. An Anton tenant cannot be reconstructed from the bundle alone: the UUIDs, the form set, the raw values and the protection periods are missing.

For the lossless route there is the native round trip (anton:export-native → anton:import-native): a package in the anton-import-format v0.10 together with master media, which Anton reads back in. Anchored via UUIDs (unknown → create new, known → skip, update with --update), hierarchy via the UUID of the superordinate unit, all languages, events, text fields of all carriers including private ones, actors as descriptors (anton#825; a re-import used to lose them), termselect values, form set and the reconnection of the media. Before every write operation the whole package is validated; if it fails, it is rolled back.

Actors, places and keywords carry a portable UUID too — a re-import anchors them to it instead of merging authority records that happen to share a name.

Since anton#706 authority records come back whole: other names, variants and abbreviations, type and subtype, the linked authority data (GND, Wikidata …), the concept schemes with URI and notation, the provisional mark, an actor's privacy and date type, a place's country and coordinates. With them the relations between them and to records. Before, these fields were missing, and private actors were lost from the export altogether, from events and descriptors alike, because the export runs on the console signed in as nobody. Backups written with anton:export-native before v0.124.0 have this gap.

Comments are lost in the round trip

«Lossless» refers to the description. Internal comments are excluded: they are kept in a table of their own that no exporter knows — not even the native one — so that they cannot end up in a publication by accident. An export-import cycle leaves them behind.

Anyone who has to take them along needs the SQL dump. That is the backup which really does contain everything.

Not included, and therefore still the SQL dump: user accounts, settings, form definitions and the AI data.

Privacy filter

RDF, TEI and EAD filter out withheld records and actors, internal-only media as well as internal text field types (internal archival remarks, cataloguer's information, comment).

EAD always shows the public view, whoever generates the file: no locked descriptions or drafts, with everything below them, and no locked actors, not even as creators. Anton serves the file of a public fonds to anybody through OAI-PMH. Up to v0.100.0 it contained what the person generating it could see, and always the locked creators (anton#616).

Plain flag logic

The filter evaluates the private flag — it is not a role check. Apart from the file name, downloading a generated RDF file checks no further permissions.

Lasting address

Since anton#784 the exports name a record by its lasting address https://<archive>/id/<uuid> instead of the page of its database number: EAD 2002 (eadid/@url, unitid/extptr), EAD3 (recordid/@instanceurl, unitid/@identifier), Dublin Core via OAI-PMH (dc:identifier, dc:relation), Memobase (sameAs), PREMIS (a second objectIdentifier of type URI) and the API (permalink, for actors too). RDF A+ and RiC-O keep their own IRIs and point to the address with rdfs:seeAlso, for records, actors, places and keywords. The zbz TEI export keeps its citation format.

Media without a record

Every export goes by record. A medium whose record has been deleted but which is still in the database and on disk is therefore in no package. So that it does not go missing silently, Anton puts a list …-media-without-record.csv (media id, file, size, checksums, path, whether the file is still there; anton#786) beside an export of the whole tenant or of a whole tree:

  • anton:export-native without --root or with a record at the top of a tree: beside the ZIP.
  • RDF bundle (a-plus-bundle) likewise, also when it is made on the export page; there the list can be downloaded from the file list.
  • anton:export-dip --format=eark --full: once per run in the target folder, <slug>-media-without-record.csv.

With no such media, no list is written. What happens to the files is the archive's decision; anton:audit-orphans --only=media shows them at any time, and the daily report anton:review-logs reports media left over after their record was deleted.

Preservation strategy (anton#807): For every decision of the preservation mandate Anton hands out the strategy document as JSON, byte for byte as it was confirmed; the header X-Content-SHA256 carries the checksum that also stands in the PREMIS rights statement. The document describes the decision, not the holdings, and is therefore no backup of the data. The confirmation itself (who, when) is held only in the database.

The essence

Descriptive metadata is excellently covered by open standards and can be packaged. Authority data has exactly one export of its own (TEI). File provenance travels as PREMIS in the DIP, OCFL and native round trip. The event graph, users, configuration and AI data have no standards-based export and can only be secured through the complete SQL dump.