Made an account just to reply to this. I do specialized programming work for Kofax Capture, KTM, etc that deal with managing scanned documents. Good documents have little indicators to help OCR align the page for recognizing characters or check boxes. I find it very hard to believe a “glitch” like this would ever happen.
Now I can’t speak for ballot software but in the suite of Kofax software you set certain thresholds for OCR to pick things up and there’s a lot of testing that’s involved with tuning everything in to give accurate readings. The indicators to align the page are a main contributor and I know ballots have these on them. And especially for an election I know for a fact these will all be highly accurate.
I’ve written many pieces of software that read zones for tens of millions of documents and storing what it picks up in databases and there is no way in hell that faults like this would ever be allowed into production.
Made an account just to reply to this. I do specialized programming work for Kofax Capture, KTM, etc that deal with managing scanned documents. Good documents have little indicators to help OCR align the page for recognizing characters or check boxes. I find it very hard to believe a “glitch” like this would ever happen.
Now I can’t speak for ballot software but in the suite of Kofax software you set certain thresholds for OCR to pick things up and there’s a lot of testing that’s involved with tuning everything in to give accurate readings. The indicators to align the page are a main contributor and I know ballots have these on them. And especially for an election I know for a fact these will all be highly accurate.
I’ve written many pieces of software that read zones for tens of millions of documents and storing what it picks up in databases and there is no way in hell that faults like this would ever be allowed into production.