Reads scanned and low-quality statements more reliably

Not every statement is a clean digital PDF. Scans, photos, and image-only files used to stall or come back empty. The extractor now works a lot harder to read them.

• More ways to read a page, with automatic fallback. The extractor now runs several OCR engines and retries a page before it gives up. If the main engine cannot read part of a statement, it hands off to a backup engine instead of failing the whole file.

• Opens files that used to break. Malformed PDFs get repaired before anything tries to read them, rotated pages are turned right-side up, and larger scans decompress instead of hitting a size limit. Statements that would not even open now extract.

• Cleaner text off scans. On scanned pages, the OCR path catches garbled or repeated output and retries it, drops repeated rows, and strips long runs of junk characters, so the transactions pulled from a scan come back more accurate.