I might be mistaken, but it seems in this new YouTube video he contradicted himself on a couple of things. Before he said the alignment of sources were performed with a semi-automatic software using the method of image registration - there even was a video demonstrating it on the SW logo. In this latest videocast he said he did the alignment manually. Also, earlier he seemed sure that he couldn’t get any additional detail from the o-neg, but now he seems to think otherwise.
Also, the difference between the raw scan of the crawl and the restored version wasn’t just ‘stacking sources’, he used a cleaner version of every line of text from earlier in the sequence to fix those deteriorated top lines.