hkdse-math-grading

Automate HKDSE Math grading by extracting rubrics, calibrating against reference data, and applying per-question mark allocations across multiple papers. This end-to-end workflow uses parallel sub-agents for per-student evaluation, outputting detailed JSON files with scores and level assignments.

Level assignment calibration in HKDSE Math grading derives empirical level boundaries from a reference cohort of 2024 level-labeled data. It analyzes ten reference students to establish baseline scores, which are then applied to assign levels 1 through 5 to new datasets.

Yes, you can grade HKDSE Math Compulsory Part Paper 1 without manual rubric entry by using automated rubric extraction. The system enforces page-by-page extraction for accuracy and applies the derived marking scheme to allocate per-question marks automatically.

Per-student JSON output for HKDSE Math scores works by compiling detailed per-question reasoning and final mark allocations into individual files. These files are saved under output/<year>/studentN.json, containing the complete grading breakdown and derived level assignments for each student.

HKDSE Math grading automation supports optional calibration and validation phases to ensure grading accuracy. Calibration derives empirical level boundaries from reference data, while validation checks the rubric application before producing the final per-student score and level outputs.