[less_less]

> An OTP must have particular properties that words in a book don’t have to be provably secure. However, common materials that the communicators have access to without specific distribution can be useful for OTPs in their own way (though not provably secure).

Using a passage from a book directly as an OTP is not secure at all. It's similar to the result of reusing an OTP: it gives an attacker two messages in (e.g.) English whose sum (using a book as an OTP) or difference (reused OTP) is known. Since the entropy of English is around a bit per character (more or less), and the deltas reveal almost lg(26) ~ 4.7 bits per character, knowing the sum or difference can determine most of the contents of the message in either case.

Reading every n'th letter in the book is probably marginally harder to break, but I still wouldn't rely on it for anything serious.

[uclibc]

Instead of using every n'th letter, how about using a pseudo random number sequence generated by a simple math equation, like the one posted here a few days ago[1], and add that to every letter. With that the data should be shuffled basically beyond recognition though the effort needed to decode has increased even more.

[1] https://news.ycombinator.com/item?id=39091867

[Izkata]

The way I've seen it used in TV shows, there'd be a number somewhere (embedded, or day of month the message was sent, etc) that indicates what page of the book to use.