category is broad and filterable, e.g. Home purchases, Business expenses, Vehicle, Tools, Medical, Travel.
doc_type distinguishes receipt, invoice, warranty_certificate, quote, etc.
Keep images immutable; fix OCR/database fields rather than replacing the source image.
Use image_sha256 to avoid duplicate imports.

Local import/OCR wrapper

# Dry-run OCR/parse without touching the DB
knowledge/receipts/scripts/receipt_import.py /path/to/receipt.jpg --dry-run --json

# Import a new receipt image/PDF into the archive and DB
knowledge/receipts/scripts/receipt_import.py /path/to/receipt.jpg --category "Home purchases" --notes "Warranty for office"

# Re-OCR/update an existing DB row
knowledge/receipts/scripts/receipt_import.py knowledge/receipts/images/26-05-images/example.jpg --receipt-id 2 --category "Personal care"

The wrapper keeps originals immutable, writes OCR derivatives under knowledge/receipts/ocr/<sha-prefix>/, uses local Tesseract/OCRmyPDF only, and stores parsed fields plus raw OCR text in receipts.sqlite.

Useful queries

-- Recent records
SELECT id, category, doc_type, merchant_name, receipt_date, total, image_path
FROM receipts
ORDER BY COALESCE(receipt_date, captured_at) DESC;

-- Filter by category
SELECT id, merchant_name, receipt_date, total, image_path
FROM receipts
WHERE category = 'Home purchases';

-- Search items/text
SELECT r.id, r.category, r.merchant_name, i.description, i.sku, r.image_path
FROM receipts r
JOIN receipt_items i ON i.receipt_id = r.id
WHERE i.description LIKE '%powerbank%' OR r.ocr_raw_text LIKE '%powerbank%';