eval-grading-quality-audit

What problem does it solve?

Automates a per-question quality audit of grading outputs by deploying a sub-agent per student to read the rubric, the extracted answer, and the grading result, scoring each question on five criteria, and reporting failures and root causes.

Core Features & Use Cases

• Per-question audit using sub-agents that independently re-judge student work against the rubric and verify consistency with the awarded marks.
• Produces grading_quality_audit.json with per-question details, including failure_stage and root_cause_type, plus a student-level summary for remediation planning.
• Integrates reference rubrics and failure taxonomy to enable traceable diagnostics and audit commentary.

Quick Start

Provide the rubric, extracted student answers, and grading outputs for the target cohort to launch per-student audits and generate the aggregation file grading_quality_audit.json.