E.11:4.6.1 - Compare the information the reader actually receives
For a title or cue revision, name the reading operation and compare the material that operation exposes. First discovery, browsing supported working questions, selecting among neighbors and re-finding a known method can need different cues. Understanding an opened body and composing its method with others require the content and relations inside the patterns.
Keep these comparisons distinct:
| Visible material | What the comparison can establish |
|---|---|
| Titles alone | Whether the headings expose a plausible working question and distinguish candidates at this first encounter. |
| Complete ToC rows | Whether titles together with keywords, query cues, status and relevant dependencies support the choice. |
| Full bodies or search windows | What the available passages let this reader recover; the result includes any body explanation actually supplied. |
| Indexed or retrieved fragments | What the consuming system can find and show from the fields and context it actually receives. |
For a title-only comparison, withhold the fuller cue and the author’s intended answer. Keep the task and allowed preparation comparable, include tempting neighboring patterns and an ordinary non-use case, then open the selected body to inspect the promised result and boundary. Preserve an actual first response before giving explanations that could teach the intended choice. A desk comparison can justify a visible repair; use an actual reading when unresolved interpretations can change the selection or when claiming that a reader recovered the intended question.
For AI retrieval, inspect the real input: the file or index fields, included heading and parent context, chunk boundary and returned passage. A renamed heading can affect only a path that carries its text or resulting representation. Adding context to a fragment is another intervention. A heading scan, a full-text search and a contextual embedding query therefore need their own declared conditions; a result from one does not establish the others.
Check both discovery from a working question and re-finding by a useful technical term. An attractive first click is insufficient if the body returns the reader from a wrong use or the familiar method becomes harder to find. Keep the useful shortlist, correction or no-use return; ordinary comparison needs no extra catalogue or score. A local AI reading supports that observed response, not a human-effect estimate. Keep any new body content separate from a title or presentation contribution, and use an appropriate description-comparison method when that attribution matters.