What problem does it solve? Manual review decisions in bibliometric screening are usually treated as one-off corrections, wasting the implicit feedback they contain. This Skill turns every confirm, overturn, or defer decision into explicit rule revisions that continuously improve the literature filtering strategy. ## Core Features & Use Cases - Structured Review Execution: Presents each record with seq, title, keywords, abstract excerpt, current verdict, and noise type, then records the user's decision and rationale back into the mapping table. - Attribution Analysis: Classifies each overturned judgment as a rule defect, boundary shift, new pattern, or isolated case, and clusters similar cases to detect reusable patterns. - Rule Revision Proposals: Generates rule revision suggestions with estimated impact scope, applies them only after user confirmation, and back-propagates changes to previously judged records. - Use Case: After a semantic screening pass flags 200 papers as noise, you review a 20-paper sample, overturn 5 verdicts, and the Skill infers that a keyword pattern was too broad, proposing a rule update that reclassifies 30 affected records. ## Quick Start Review the PENDING literature from the last screening batch and update the filtering rules based on my decisions.
