Sources
10 pages filed under Ingestion › sources. A term is a word of the business with its definition, its aliases and the words it relates to.
| Term | aliases | First line | Links |
|---|---|---|---|
| Corpus | knowledge corpus | The notes and documents of every source a build reads together: what the dictionary, the search index and the alphabetical index cover, and what a candidate expression recurs across. A golden corpus is a fixed corpus with the model the engine must produce from it. | 97 |
| Fingerprint cache | conversion cache, pipeline cache | The folder, .concordance-cache by default (conversion.cache), where the build keeps what it fetched or produced, keyed by the fingerprint of its input: a converted document under convert/<sha256 of the file>, so that an unchanged document is never reconverted even when it moves, and a contract by the SHA-256 of its text. The clones of the sources sit under it too. The cache never enters dist/; a pipeline keeps it between runs and deleting it forces a full reconversion. | 30 |
| Knowledge repository | repository, content repository | A git repository, or a local folder, that holds notes and documents and is declared as a source of the wiki. The tool never writes into it, each knowledge repository lints itself in its merge requests with the linter, and this glossary and the specifications of the tool are two of them. A repository that holds the configuration and no content is the configuration repository. | 148 |
| Privacy | privacy block | The privacy: block of the configuration: exclude, the globs of what is never read, applied before any content (an excluded file, like the ones a repository excludes itself or git ignores); pseudonymize, the pseudonymisation of transcripts with its dictionary and scope; and publish_transcripts, false by default, so that a transcript is indexed only when a decision asks for it. The configuration never carries a credential either: a private source gets its token from the git environment of the machine or of the pipeline. | 35 |
| Pseudonymisation | pseudonymization, pseudonymise | The replacement of the speaker names of a transcript, and of the personal mentions detected in it, by stable pseudonyms before anything else reads it: privacy.pseudonymize in the configuration, off by default, with the types concerned, the dictionary pseudonyms.yaml of real names to pseudonyms that is never published, and whether a known role stands in for the pseudonym. A name pattern found outside the dictionary yields I-PII-DETECTED for review. Pseudonymisation is a mechanism, not a permission: transcripts are published only when privacy.publish_transcripts asks for it. | 60 |
| Representation | form of a document | One of the files an entity is made of: its note, a slide deck, the PDF a converter produced from it, a transcript, the contract an operation was imported from. A note alone has one representation; twin resources merged into one entity carry several, and its page shows them one after the other, the file for download, its extracted text and its viewer. An office document without markdown next to it is reported by W-DOC-NOMD until a note becomes its representation. | 67 |
| Source | source repository, declared source | A git repository, or a local folder, declared in the configuration with its typing rules. Every entity's identifier starts with the name of its source, and every occurrence knows which source it came from. | 174 |
| Staleness | stale, dormant source | The age after which a source or a note is flagged W-STALE: staleness.warn_after_days in the configuration, a default and per-source overrides, the dates coming from git. The home page reads the same thresholds and marks a source dormant in its freshness entry when its newest change is older than its threshold at the instant of the build, 180 days without the key. | 53 |
| Twin resources | twins, representations of a document | Several files that are forms of the same document: a slide deck, its markdown notes, the transcript of the meeting where it was shown. The tool scores their likeness (same base name, same title, similar text, same commit, or an explicit declaration) and merges them into a single page when the score is high enough. Below that, it reports a candidate and waits for the lock. | 89 |
| Version | schema version, version key | The first key of every file the tool reads, version: 1 in this version: the configuration, the profile, the theme and the lock each name the version of the schema they follow, and a file of another version is refused with the path of the key. The tool itself carries a version too, written in the footer of the site and in the build block of the model, and every published package shares it. | 8 |
10 sources of 10 — pagination by twenty. The Links column counts the related pages, which brings the most central sources of the journey to the top.