Research guide
Tips for theological and literary research. Search effectively, understand similarity scores, and export your findings.
Writing effective queries
Golden Thread uses semantic search: it finds passages by meaning, not keywords. This changes how you should search.
Describe the concept, not the term
Ask questions, not keywords
Try multiple formulations
Semantic search finds different passages depending on how you phrase the question. For thorough research, search the same concept in 2–3 different ways and compare results.
Understanding results
Similarity scores
Press Shift+Ctrl+D to open the debug panel, then enable "Show internals" to see how closely each result matches your query. Scores range from 0 (no match) to 1 (perfect match).
- 0.85+ — Very strong semantic match
- 0.70–0.85 — Good match, worth reading
- 0.55–0.70 — Tangential, may have relevant context
- Below 0.55 — Weak match, likely noise
Document IDs
Each result has a stable document ID visible when the debug panel is open. IDs follow the pattern collection/source/identifier. These IDs are permanent and can be used in citations.
Collection coverage
| Collection | Sources | Status |
|---|---|---|
| Scripture | Douay-Rheims Bible | Complete |
| Catechism | CCC, Baltimore Catechism | Complete |
| Councils | Vatican II documents | Complete |
| Fathers | Apostolic Fathers, Internet Archive patristic corpus | Growing |
| Literary | Chesterton, Belloc, Hopkins, Thompson, Newman, Pascal, Dante, Milton, Chaucer | Growing |
| Theology | Summa Theologica, theological works | Growing |
| Artists | Living Catholic artists (seed data) | Seed |
Limitations to know
- No full-text — Golden Thread indexes chunks (paragraphs or short passages), not complete works. Use "Read at source" to see the full context.
- No cross-references — Results are independent passages. We don't yet link related passages across collections.
- No keyword search — If you need an exact phrase match, this tool searches by meaning. A passage must be semantically close, not just contain your words.
- Chunk boundaries — Passages may start or end mid-thought due to chunking. Always click through to the source for full context.
- Embedding model — We use
multilingual-e5-small(384-dimensional, 100 languages). Cross-language search is supported — you can query in English and find Latin, Greek, or other-language passages. Results for highly specialized terminology may be less precise than for natural-language queries.
Export and citation
Use the ☆ Pin button to collect results, then export as CSV or Markdown from the partner workspace (available with partner authentication). Each export includes document IDs, citations, source URLs, and passage text.
Start searching →