benchclaw-stage4-cdm-irt-analysis

What problem does it solve?

This Skill addresses the challenge of performing item response theory (IRT) and cognitive diagnosis model (CDM) analyses on grey-batch validation results, allowing for detailed diagnostic checks on the performance of a model's ability.

Core Features & Use Cases

• IRT/CDM Diagnostics: Provides analysis of score matrices from grey-batch validations to estimate item difficulty, discrimination, and model ability.
• Subskill Activation: Designed as a subskill within the larger BenchClaw benchmark pipeline, requiring explicit dispatch via /benchclaw-subskill command.
• Use Case: In the context of a benchmark's grey-batch validation phase, use this Skill to diagnose issues in small-batch model responses and ensure data quality before full synthesis.

Quick Start

To execute the analysis, ensure you're running in the appropriate sub-agent context, then invoke the skill with /benchclaw-subskill <SKILL_PATH> <Skill_Name> <Project_Root> <Node_ID> <Input_Artifact> <Output_Artifact> <DAG_Dependency>.