Lesson 10: Indexes, Top-k, and source information
Three search results are not an answer yet. Ask how they were selected and ranked. Indexes and Top-k explain what that list actually represents.
An index organizes lookup
A book index links terms to pages; a search index links searchable features to source locations. Keyword and vector indexes differ but both make candidate lookup efficient. Results need text and provenance so they can be inspected.
After a document update, check its index too. If the source says three but retrieval supplies an old two-book passage, editing visible text did not update the whole chain.
Top-3 means the first three candidates
The k is a result count. Top-3 selects leading candidates under the retrieval or ranking method. It guarantees neither relevance nor correct claims, and filters or limited candidates can return fewer than k.
For current limits, simulate E/B/H. Top-1 contains the old rule; Top-2 adds current B. H concerns reference-book restrictions rather than ordinary borrowing quantity.
For special versus ordinary opening, A/H/D misses D at Top-2. This query needs the notice, but that does not establish a universal three-result setting.
More results, or better ordering?
Too few candidates can miss evidence; too many add noise, repetition, and burden. Check necessary evidence, then consider count and order. Later evaluation treats them separately.
Keep IDs, titles, dates, status, and origins. They enable citations, version decisions, and source review. Scores can assist without replacing provenance.
Create eight cards with answerable questions. Compare first one, two, and three results in both lists and mark missing evidence. Manual lists are exercises; real evaluations need the actual returned order.
What should change
Current limits gain B at Top-2 but still need archive checking. The date comparison gains complete A/D at Top-3. Whether more results help depends on useful evidence versus noise.
Further reading: Microsoft: Query a vector index
← Previous · Learning plan · Fictional library practice sources · Next →