What problem does it solve? After each round of bibliometric noise removal, researchers need consistent, verifiable deliverables: a cleaned dataset in the original format, a traceable exclusion list, and an updated cleaning log. Doing this manually risks record loss, format corruption, and untraceable decisions. ## Core Features & Use Cases - Round-based export: Generates a retained-records file (preserving the original WoS plain-text or CSV format) plus an exclusion CSV listing seq, source_id, title, noise type, reason, and round for every record judged as NOISE. - Final dataset assembly: Builds the final deduplicated dataset directly from the original raw data using the final mapping table, preventing field-integrity loss from intermediate files. - History traceability: Maintains verdict_history, per-round noise_type fields, last_verdict, and first_round columns so every exclusion is auditable. - Validation and logging: Enforces record-count conservation checks and updates the cleaning log with noise-type distribution tables and output file inventories. - Use Case: After S07 finishes semantic judgment for round 2 of a Web of Science literature cleaning project, run this skill to export the retained WoS file, produce the round-2 exclusion CSV, and append the round summary to the cleaning log. ## Quick Start Run the round-2 export now: generate the retained records file and exclusion list from the current mapping table, then update the cleaning log.
