Namespace doc: is https://ns.flur.ee/doc#. Structure elements additionally use DoCO (doco:, http://purl.org/spar/doco/), the pattern ontology it extends (po:, http://www.essepuntato.it/2008/12/pattern#), NIF (nif:, http://persistence.uni-leipzig.org/nlp2rdf/ontologies/nif-core#), Dublin Core (dcterms:, http://purl.org/dc/terms/) and FOAF (foaf:, http://xmlns.com/foaf/0.1/) for the document root’s class. The structure terms are those of the parsing engine and are documented in full in its vocabulary reference; the terms below are what the ingest adds.
Path relative to the ingested directory, /-separated.
doc:sha256
string
Hex digest of the bytes.
doc:mediaType
string
The media type inferred from the extension.
doc:byteSize
integer
Size in bytes.
doc:pageCount
integer
Pages, for sources that have them; 0 otherwise.
doc:escalatedCrops
integer
Crops a vision model read for this extraction.
doc:parserRevision
string
The fluree-doc-parse revision that produced the structure graph.
doc:chunkCount
integer
Chunks written.
doc:chunking
string
min/max characters the chunks were cut with.
doc:embeddingModel
string
The embedding model, when one ran.
doc:embeddingDimensions
integer
Its vector width.
doc:ingestedAt
xsd:dateTime
When this extraction was written.
doc:extractionModel
string
The language model that extracted, when one ran.
doc:extractionFingerprint
string
Hash of the ontology, entity sources, model, guidance and relation mode.
doc:mentionCount
integer
Mentions written.
doc:entityCount
integer
Distinct entities mentioned, known and new.
doc:relationCount
integer
Relations written, every verdict included.
doc:readerPredicate
IRI
A predicate a reader wrote outside the pipeline’s own; a re-ingest retracts these from the nodes the reader minted, and nothing else there.
doc:readingGraph
IRI
The named graph an HTTP reader’s reading of the document is committed in (<document>/reading), which a re-ingest clears before it commits the new reading. The reading itself is in the extraction vocabulary, xt: (https://ns.flur.ee/extract#).
The statement’s object: an entity, or the literal the model gave when the object was a value or named no entity in the chunk.
rdfs:label
string
`subject
doc:excerpt
string
The text supporting it, as the model quoted it.
doc:verdict
string
valid, repaired or rejected.
doc:assertionMode
string
asserted, hedged, attributed or negated: how the source states it. Present only when a custom --system-prompt asked for it. Not a confidence score and not ordered — see Assertion mode.
doc:asserted
boolean
Whether a direct edge was written for it.
doc:originalPredicate, doc:repairNote
string
On a repaired relation: what the model wrote, and how it was resolved.
The chunker starts at the foaf:Document root and follows po:contains, visiting a node’s po:containsAsHeader child first. It takes text from doco:Title and doco:Paragraph (nif:isString, else rdfs:label), turns each doco:Section’s doco:SectionTitle and doc:sectionLevel into the header path, and reads each doc:TableCell as doc:cellValue with its doc:rowHeader and doc:columnHeader. Every other node is a container it walks through. See What gets built for the rest of the structure graph.