Top-k sets how much the model reads.
Twelve Larkspur Cloud handbook chunks, indexed with real TF-IDF as in lesson 10.2. Pick a question and move k: the chunks above the magenta rule are what the LLM receives. A chunk that scores zero stays out at any k.
- Recall@k
- Precision@k
- Context sent, about this many tokens
- All evidence retrieved
Ranking for this question
Cosine similarity, TF-IDF
| Rank | Chunk | Score | Golden set | Text |
|---|
Rows at 0.000 share no term with the query. A real index returns them in arbitrary order, so the lab never counts them as found. Tokens use the four-characters-per-token rule of thumb.
The baseline row at this k
| Golden question | Recall@k | Precision@k |
|---|---|---|
| Mean |
This is the row the M10 lab asks you to record. Every later module (better chunks, embeddings, hybrid search, reranking) reports its change against a row like this one, on the same questions.