A parting thought. The scenario that you propose is perfect for A/B testing if you have the traffic volume to support it. Testing different reCAPTCHA treatments head-to-head would give you much clearer data than fallout analysis of different treatments in different timeframes.
The least that i want to track- when a user completes the action/reCaptcha puzzle/question and hits Done or 'am not a robot' radio button.
Why- We want to know how user flows gets impacted(across a set of pages/section) or not when we change the severity of the puzzle and also get a better idea in terms of what % of users are actually getting and attempting the reCaptcha. Thank you for looking in to this for me.
This would be pretty difficult to do, as I believe those are loaded in frames. However, if recaptcha fires off some sort of event, you might be able to listen for it and send it in an Analytics call. Can you share more on what you'd like to track in reCaptcha and why?