Homework Grading Workflow

Process scanned homework: extract names from pages using vision → match to roster → create individual PDFs → update completion spreadsheets.

Core accuracy rules: Read EVERY page individually (no batching) | Focus on Name field at TOP | Verify EACH PDF after creation

Required inputs: Scanned homework PDF + Student roster spreadsheet (with period sheets) + Completion spreadsheet (optional)

Workflow Overview

Phase 0: Planning (BEFORE ANYTHING ELSE)
├── Count pages in PDF (fitz.open → len(doc))
├── CREATE tracking file with batch plan
├── Calculate sessions needed: ceil(total_pages / 99)
├── Display plan to user:
│   "150 pages = 2 sessions (99 + 51)"
└── User confirms before proceeding

Phase 1: Setup
├── Load roster from xlsx skill
├── Extract PDF pages as images
└── Tracking file already exists from Phase 0

Phase 2: Page Analysis (CRITICAL)
├── Check tracking file for current_session and start_page
├── Read EACH page individually (max 99 per session)
├── Find Name field at top
├── Match to roster (fuzzy matching)
├── Save to tracking file after EACH page
└── Stop at session limit, inform user to resume

Phase 3: User Verification
├── Display uncertain pages
├── User confirms/corrects names
└── All pages must be assigned

Phase 4: Create PDFs
├── Group pages by student
├── Create individual PDFs
└── Skip Unknown pages

Phase 5: Update Spreadsheet
├── Add assignment column
├── Mark submissions with X
└── Update Total formula

Phase 6: Verification
├── Check each PDF has correct pages
├── Cross-reference spreadsheet
└── Cleanup temp files

Quick Reference

Phase	Reference File	Script/Template
Decision Flow	workflow-flowchart.md	-
Page Analysis	page-analysis.md	scripts/extract_pages.py
Status Tracking	status-tracking.md	scripts/update_status.py, templates/homework-grading-status.yaml
PDF Creation	pdf-creation.md	scripts/create_student_pdfs.py
Spreadsheet	spreadsheet-update.md	-
Verification	verification.md	-
Troubleshooting	troubleshooting.md	-
Multi-Model Analysis	clink-integration.md	PAL MCP clink tool

Required Skills Integration

Always invoke these skills:

Skill: document-skills:xlsx  # For spreadsheet operations
Skill: document-skills:pdf   # For PDF manipulation

Critical Rules

Phase 0 (BEFORE ANYTHING ELSE):

• Create tracking file FIRST → Calculate batch plan (ceil(pages/99)) → Display plan → Get user confirmation

Page Analysis:

• ONE page per Read call (never batch) | Focus on "Name:" at TOP | Save to tracking file after EACH page | Stop at 99-image limit

Verification:

• Open and verify EACH student PDF | User MUST resolve uncertain pages | Cross-check spreadsheet X marks

Red Flags → STOP

If you see...	Do this instead
No tracking file created	Create tracking file FIRST with batch plan
Batching pages	Read ONE page per Read call
Unknown pages being processed	User MUST confirm names first
Skipping PDF verification	Open and check EACH PDF
Not saving after each page	Save to tracking file after EVERY page

Confidence levels: high (exact match) → proceed | medium (fuzzy match) → proceed with note | low/unknown → flag for user review. See page-analysis.md for details. For difficult handwriting, use clink to get second opinion from another model - see clink-integration.md.

Session Resume

99-image limit per session. Tracking file ({output_folder}/homework-grading-status.yaml) enables automatic resume. See status-tracking.md for schema and resume workflow.

Output Structure

{output_folder}/
├── homework-grading-status.yaml
├── Student Individual Files/{Student}.pdf
└── {completion_spreadsheet} (updated)

Prerequisites

pip install PyMuPDF pandas openpyxl pyyaml filelock