Dedupe + Rank

Turns a raw candidate pool into a deduped pool and a stable core set.

Input

• papers/papers_raw.jsonl

Outputs

• papers/papers_dedup.jsonl
• papers/core_set.csv

Script boundary

scripts/run.py should own only:

• title/year deduplication
• deterministic ranking
• stable paper_id generation

Use shared domain packs or pipeline contract metadata for topic-specific or product-specific behavior.

Contract-driven behavior

The script should prefer pipeline contract metadata over profile-name branching.

Current important field:

• quality_contract.candidate_pool_policy.keep_full_deduped_pool

If true, the script keeps the full deduped pool in papers/core_set.csv unless the user explicitly overrides core size.

Acceptance

• deduped JSONL exists
• core-set CSV exists
• reruns are stable for the same inputs

Non-goals

• retrieval
• screening
• manual topic authoring inside the script