# Taxonomy Guidelines

Use these rules to keep the classification defensible and useful for the review.

## Priority order

Prefer classification by:
1. Research task
2. Method family
3. Application domain

Only prioritize application domain when the user explicitly wants an application-focused survey.

## Design constraints

- Use one consistent basis across the whole corpus.
- Keep category names short and academically natural.
- Avoid mixed-level labels such as one category by task and another by dataset.
- Avoid vague buckets such as "other" or "miscellaneous" unless the corpus truly requires one residual bucket.

## Corpus size guidance

- 1-3 papers: keep the taxonomy shallow and note that the grouping is provisional.
- 4-12 papers: 2-5 categories are usually enough.
- Larger corpora: add one subcategory layer only when it clearly improves the review.

## Evidence order

Use these sources in order of trust:
1. Extracted abstract
2. Extracted introduction and method snippets
3. Extracted evaluation snippets
4. Title and metadata

If extraction quality is weak, reopen the raw source selectively instead of pretending the evidence is stronger than it is.