# Module 4 practice · explain a ranking

Attempt using [the lesson](../lessons/04-explain-a-lexical-baseline.md), then read [the solutions](../solutions/04-lexical.md).

**A. A different query.** Reuse A=`red blue` and B=`red`, but query `red`. Compute both cosine scores using the lesson's normalized vectors. Which passage ranks first? Why does the longer document not automatically win?

**B. An absent concept.** Compare the queries `red` and `red laptop` in that same tiny index. Predict whether their scores differ. Run the comparison using `Index` and `Document`. Explain why this behavior is dangerous if a user asks about a specific object unknown to the collection.

**C. Cutoff and completeness.** Run `methods.py lexical`. Find `q-two-sources` in each report. Compute its recall at cutoffs 1 and 2, then explain the overall means `0.8125` and `0.875`. State why mean reciprocal rank does not change. Do not use no-evidence questions in the recall denominator.

**D. A meaningful check.** Describe a test for a future change that starts indexing passage titles. Which current code/test assumption would need updating? Design one tiny corpus/query where title indexing changes a result, without adding a live service or calling the change an improvement before evaluation.

Self-review: show calculations, identify dropped unknown terms, use the eight-case denominator and state precisely what the proposed test verifies.