Documents

14 pages filed under Ingestion › documents. A term is a word of the business with its definition, its aliases and the words it relates to.

TermaliasesFirst lineLinks
Conversionoffice conversion, conversion blockThe transformation of an office document into a PDF at build, by a converter plugin, so that its text is extracted from that PDF and nothing else, and the conversion: block of the configuration that bounds it: timeout_s (120) per document, max_size_mb (50) beyond which a file stays unconverted, cache, the fingerprint cache outside dist/, and parallelism, the number of concurrent conversions, which changes the build time and never the output. A document that fails to convert yields W-CONV-FAILED and stays downloadable; no converter runs in this version.
Converteroffice converterA contribution point of the plugin API that turns a document into representations the pipeline reads: a PDF today, thumbnails and text later, written under the fingerprint cache and never next to the source. A converter receives the bytes of the file, their SHA-256 and the limits of conversion: (timeout_s, max_size_mb), runs in a pool of conversion.parallelism workers, and reports a failed conversion as W-CONV-FAILED, the document staying a download. The official converter drives headless LibreOffice; no converter runs in this version.
Cuetimed cue, timecodeOne timed unit of a transcript: a start and an end timecode and the words spoken between them. A cue is the position of a word read in a transcript, as a page or a slide is in a converted document: a mention cites it by its timecode, and the rail of the document viewer lists the cues with the first line of their text.
Deckslide deck, slides, slideA slide presentation, a .pptx file that a converter turns into a PDF at build so that the site previews it and reads its text slide by slide. Its page offers the original for download, the PDF the conversion produced or the preview committed next to it, a rail of its slides captioned by their first line and the extracted text of every slide; a mention read in a deck cites its slide. A deck is often a twin resource of the notes and the transcript of the session where it was shown.
Documentraw document, untyped resourceA resource that no rule types: a markdown file left as-is, a slide deck, a Word file, a PDF, a transcript. A document is indexed and its words are recorded from its extracted text, but it enters the model only through what it documents. Its page offers the file for download, shows its extracted text page by page and, when the conversion produced a PDF, opens it in a viewer on demand.
Extracted texttext of a document, extracted pagesThe words of a document as the build reads them: the text of every page of the PDF the converter produced, or of the PDF given as a source, and the cues of a transcript with their timecodes. An office document never gives its text any other way, so that one extraction path serves every format. The extracted text feeds the search index, the recognition of occurrences and the comparison of twin resources, each word keeping the page, slide or cue it was read on, so that a mention cites slide 3 where a note cites a line. The page of the document shows it position by position, cut at the configured size; the download gives the rest.
FrontmatterYAML frontmatterThe YAML block between two --- lines at the top of a note: the common attributes every entity may declare (id, type, title, aliases, application, domain, status, tags, summary, superseded_by, locale) and the attributes of its type, a frontmatter reference among them. The frontmatter is the last rung of the type cascade, is never scanned for occurrences, yields E-FM-INVALID when it does not parse and W-ATTRIBUTE-UNKNOWN for a key nothing declares. What qualifies goes in it; the rest is prose.
MarkdownCommonMark, GFMThe only syntax the tool reads: CommonMark, the tables and task lists of GFM, and an optional YAML frontmatter. No wikilink, no directive, no proprietary tag: every note stays readable without the tool, and structured information goes through the frontmatter, the sections and the lists. A note is rendered to sanitised HTML at build, raw HTML dropped, its written links turned into page links.
Meetingworking session, session, minutesA working session the documentation keeps as a note typed meeting: its minutes, with the date, the participants and the decision the session produced under decisions, the transcript of what was said and, when one was shown, the deck, grouped by the build as twin resources into one page with a tab per file. A meeting brings passages, not concepts: it enters the model through the pages it mentions and through the decision someone wrote as a note of its own, which the page names in a callout.
Notetyped note, ficheA markdown file that the type cascade recognises as a typed entity: a screen, a rule, a business object, an API, a process, a decision, a term. A note has a page made of its rendered markdown first, at most five declared properties, then the occurrences found elsewhere.
PreviewPDF preview, previewsThe PDF form of an office document, produced by the conversion at build and cached by fingerprint, or committed next to the original by its author; either way the page of the document opens it in the viewer on demand and the original stays downloadable. previews: false on a source keeps the PDFs out of the published site: the download link and the extracted text remain. A committed preview shares the base name of its original and repeats its text, two of the signals that make them twin resources.
Readerformat readerA contribution point of the plugin API that reads a file format: it receives the path and the raw bytes of a file and returns its native metadata (title, author, dates, counts) and its text, the cues of a transcript for the VTT reader. A reader is synchronous and pure, reads nothing but the bytes it is given, and throws a plain error naming the file when it cannot, which the pipeline turns into a finding.
Sectionnote sectionA part of a note under a heading of level one or two: the fragment of the note holds one rendered section per heading, every occurrence records the section it was read in, and a section whose heading the profile maps to a relation turns the notes named in it into section mentions. Headings are matched without regard to case or accents, in the language of the source.
Transcriptmeeting transcript, VTT fileThe timed text of a meeting, a .vtt file that a reader turns into cues with their timecodes. Its words enter the search index and the occurrence scan once pseudonymisation has run, a mention read in it cites its timecode, and it is often a twin resource of the deck and the notes of the same meeting. A transcript is indexed only when the configuration publishes transcripts explicitly.

14 documents of 14 — pagination by twenty. The Links column counts the related pages, which brings the most central documents of the journey to the top.