paper-triage

Paper Triage

Turn an ambiguous paper query into a ranked shortlist and clear next steps. Use this when the human starts with a fuzzy title fragment, overloaded term, or vague memory of a paper.

Arguments

• The positional argument is the paper query. Quote multi-word titles when running from a shell.
• --shortlist-size <n> controls the final shortlist size returned to the human.
• --relevance-limit <n> controls the first relevance-search pass.
• --bulk-candidate-limit <n> controls the wider recall pass.
• --snippet-candidate-limit <n> controls how many preliminary candidates get snippet search.
• --snippet-limit-per-paper <n> caps snippet evidence gathered for each snippet target.

Workflow

1. Run python scripts/run.py ....
2. Read result.possible_interpretations to see how the query was resolved.
3. Read result.shortlist for the ranked candidate papers.
4. Read result.follow_up_actions to know which workflow to run next once the right paper is identified.
5. If the shortlist still looks wrong, rerun with a more specific query string.

Output

• The script prints the unified JSON envelope described in output_contract.md.
• The underlying workflow result is TriageResult.to_dict().
• result.notes captures extra execution notes, and result.considered_candidates shows the rerank breadth.

When To Escalate

• The query is too vague to resolve into a useful shortlist.
• The top interpretation is clearly wrong even after adding specificity.
• The human really wants author lookup or direct paper details instead of title disambiguation.