> This still means that when presented an entirely new CAPTCHA, a human won't get accepted as one even if he/she answers the CAPTCHA correctly, because the reCAPTCHA system doesn't yet know he/she did so.

Well that's the whole question here. If you have a new captcha image, you can't use it for validation. If you have a captcha image where you've already amassed enough replies to be reasonably certain of the correct answer, you no longer need to continue to use it for converting the scanned image to ASCII.

So I still don't understand how this idea works for both validation *and* conversion.

You're saying that if there's a new captcha image, it rejects your answer, no matter what it is? Then, presumably, it presents a second captcha image, this time one with a history of responses behind it, and if your answer agrees with some minimum number of prior responses, you're validated as a human?

I guess that would work, except that it would require people to pretty much always enter at least two different captcha interpretations, the first being a new image that always rejects you, and the next being one that could potentially validate you and in any case help confirm the correctness of an interpretation supplied by an earlier user.

Huh.