Seeing redacted docs like this always makes me wonder - for brief blacked-out passages, couldn't you make measurements of the remaining letters/words on the line, their sizes and spacing, and algorithmically generate a few likely candidates for the blacked-out text?

You could at least estimate the length in characters of the blacked-out text. For a monospaced font this character count is trivial; for a proportionately-spaced font it'd be a little harder but you have lots of other non-censored characters to learn from.