subject-grading-v2

Automated grading of anonymized PDFs is achieved by validating student answers against a structured rubric schema to generate per-question and total scores. The workflow consumes structured questions.json and rubrics.json files to produce per-student JSON outputs and aggregated results.

Rubric-driven assessment works by loading questions.json and rubrics.json to evaluate anonymized student submissions against defined criteria. The process validates data invariants, renders deterministic grading guides, and outputs per-student JSONs plus aggregated results across multiple subjects.

Reference data can be used for score calibration across different cohorts by leveraging optional reference inputs for level calibration. Configurable settings for year, subject, and LEVEL_COUNT allow mapping scores to levels to ensure comparability across cohorts.

The best way to grade multiple subjects is running an end-to-end workflow orchestration from extraction to final scores. By consuming structured v1-schema questions.json and rubrics.json, the system processes multiple students and subjects to produce aggregated grading results.

Automated grading of anonymized PDFs requires submissions to be anonymized prior to processing. The Skill operates across multiple subjects by consuming structured questions.json and rubrics.json to validate data invariants and render deterministic grading guides.