Search every abstraction level.
Compare source passages with a mixed-level index. Scores use real TF-IDF; summaries are authored, representative fixtures.
- Positive-score hits
- 0
- Summary hits
- 0
- Source identifiers reached
- 0
- Context words
- 0
Choose a question and candidate pool.
Ranked nodes
Shared TF-IDF vocabulary
| Node | Level | Score | Context |
|---|
Zero-score nodes stay out. Ties use fixture order. Level 0 is source text; higher levels are summaries.
Selected context
Source identifiers describe ancestry. They do not prove that a summary preserves every source fact. This fixture compares candidate pools, not model answer quality.