What problem does it solve? It turns unstructured theorem-level gaps found in a machine-learning literature survey into a small, de-duplicated set of normalized research perspectives, so downstream idea generation starts from coherent, gap-supported theory lenses instead of ad hoc brainstorming. ## Core Features & Use Cases - Gap-to-perspective normalization: Reads LITERATURE_SURVEY.md and maps each retained theorem-level gap into one complete tuple of analysis_target, model_class, data_assumption, regime, and algorithm. - Duplicate control and prioritization: Merges exact and near-duplicate tuples, rejects weakly supported candidates, and caps output at MAX_PERSPECTIVES in priority order. - Checkpoint rerun support: Incorporates pending feedback from workflow_feedback.md when rerunning the perspective-selection checkpoint. - Use Case: After surveying optimization theory literature, run this step to produce Perspective_Selection.md listing three distinct, gap-traceable perspectives that branch-local idea generation can build on. ## Quick Start Run the perspective-select step on my LITERATURE_SURVEY.md to generate a normalized Perspective_Selection.md with at most three prioritized perspectives.
