Library / Knowledge-Corpus Access Engineering Principles Framework
Jump to passage
In this reading

Link to current text

Published source confirmed at last check

Source changed 2026-10-03 10:39:28 UTC · snapshot created 2026-10-03 10:40:04 UTC · last check 2026-10-03 11:30:17 UTC

KCAE.Preface:3 - The objects an engineer must keep distinct

A source item is a particular document, recording, data release or other authored or observed material. An edition identifies its particular content. A corpus snapshot fixes membership and the selected editions. A reading unit is the context needed to understand a contribution; it can span an enclosing section, definition, table note and another document. A retrieval unit is the unit indexed or ranked for discovery. A small retrieval unit may point to a much larger reading unit.

A derived view includes a text extraction, inverted index, embedding collection, graph, summary, card or transformed package. Each has a source basis and an obtaining procedure. A route is an available operation for reaching material, such as exact reading, ranked search, link traversal or direct block inspection. Several routes can share a blind spot, especially when all were built from the same summary. Complementarity means that their different selection grounds can expose different material; it is an engineering hypothesis to test.

A candidate contribution is something a source might supply to the current work: a fact, distinction, explanation, procedure, criterion, example or input for another operation. An assessment says what the inspected material supports under stated conditions. A recommendation adds that pursuing the contribution is worthwhile now. An application result is the actual explanation, calculation, decision or changed work obtained using it. Returning a document identifier performs none of the later operations by itself.

Keep four clocks where the use needs them: the source’s edition/publication time, the period in which its rule is effective, the time at which the system observed it, and the time of the receiving action. A newly uploaded file can describe an old rule. A preserved historical rule can be correct evidence for an incident and incorrect instruction for today’s operation. Authority is a separate relation: a local source owner determines which publication governs which class of decisions. File recency and retrieval rank cannot supply that determination.

The access arrangement is actual software and human work. A stored assessment is a description, not an executed repair. A skill contains instructions, not a continuously operating observer. A package version identifies distributed content, not proof that every client loaded it. These distinctions change architecture choices, so they remain visible in interfaces and examples rather than being confined to terminology.