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 11:25:15 UTC

KCAE.Preface:2 - Why changing questions defeat fixed representations

An index is a purpose-selected view of source material. Its terms, vectors, extracted entities and summaries retain distinctions that help some operations and remove others. If two different passages obtain the same derived representation, a future question can depend on the difference between them. Searching only that representation cannot recover the difference. This argument concerns a lossy mapping and an unrestricted future question; it does not imply that all compression is lossy or that a well-chosen index is poor for its intended workload.

Keeping the original creates a possible second route. It helps only if the reader can reach the relevant original independently of the representation that omitted it, afford the inspection and interpret the result. A source link behind an exclusive failed category filter leaves the practical problem intact. Conversely, reading all source bytes does not establish that a person or model noticed all relevant relations. Coverage of bytes, semantic recognition and justified use are different claims.

Questions also change through inquiry. A request for more training may become a question about an unavailable approval, then a question about a shared resource. Preserve the original episode and distinguish a new question from a synonym added for retrieval. Search translations and hypothetical answers help discover candidates; they do not become case facts. This field therefore combines prepared views with query-time interpretation and routes back to sources. The balance depends on workload and available readers, rather than on a universal preference for either precomputation or agentic exploration.

This distinction has a long information-retrieval lineage. Bates’s 1989 account treats searching as an evolving inquiry that collects contributions through several techniques. It is a historical source for preserving the developing question, not contemporary evidence that one software implementation performs best. KCAE.SEARCH gives the present engineering construction; KCAE.Profiles compares current implementation choices.