classify-capability

To classify robotics research papers by demonstrated capability, you can apply heuristic checks to single-paper inputs containing task chains and reported results, assigning each paper to a reliable, sometimes, or can't-yet tier for frontier mapping pipelines.

Criteria for reliable versus can't-yet robotics capability tiers include evaluation realism, diversity of objects and environments, success-rate thresholds with confidence intervals, independent reproduction status, and guideline overlap assessed from the paper's reported results.

Evaluating evaluation realism and success rates involves applying ordered heuristic checks on reported_results data, specifically analyzing real_robot versus simulation environments, success rates, n_objects_tested, and perturbation_tested fields to determine capability confidence.

Yes, you can process multiple robotics papers in a loop to aggregate demonstrated capabilities, as this classification function is designed to be called repeatedly by frontier-mapping pipelines to rapidly and conservatively triage many papers for research prioritization.

The paper capability classification produces a structured JSON output containing the assigned tier, capability_description, evidence list, rationale, confidence score, and boundary_notes to enable consistent downstream aggregation and survey synthesis.

Required input fields to classify a research paper's capability tier include task_chain, reported_results with success rate and n_objects_tested, environments, real_robot status, perturbation_tested, ablation_strong, venue, year, and domain.