Confirmed — every site with ReCaptcha difficulty at some higher levels needs up to 5 screens and 25 clicks till one finally reaches the 'confirm' button. And more often than not after clicking it the sad message is 'Please try again'. Without any single hint or clue why the former clicks should be wrong.
Possibly due to the facts of 'only 1 error allowed' plus:
a) Lousy image indexing. There is no defined percentage of area above which a square should be checked. Sometimes small 'street signs' e.g. are simply forgotten / overlooked by the indexer. A sidepost's rim often doesn't count, even it it's 7% or so of a square. A squire with less than 5% often counts. No algorithm whatsoever seems to be published. Of course nor, at it would be more easily implemented by a computer than by a human.
b) Incoherent image indexing. E.g. sometimes a horizontal warning sign at the road side ( [ >>> ] black on yellow or red on white) counts as 'street sign', sometimes not. Sometimes only traffic signs are asked for, sometimes any ad some enterprise has put up.
c) Imprecise translations. E.g. a 'Limousine' in German is any (4 door) car with solid top, no station wagons, no drop-tops. In English ReCaptchas it means only the 6+ Meter sedans.
Summarized this picture is a shame for a big player like Google. It hurts site visitors, wasting their time and straining their nerves. It hurts site owners, as it drives away prospects.
Honestly, who of you guys who ushered this glorious (...) would him- or herself spend four minutes on _one_single_ ReCaptcha? And how often would you try before looking for another site offering comparable goods or services without harassing you during login?
Awhile ago SolveMedia annoyed everyone with completely undecipherable, very distorted non dictionary CAPTCHAs. It cost them at least 20% of their customers; most of them turning to ReCaptcha. But hey, SolveMedia still exist. And their CAPTCHAs are readable those days.
No harm meant. A wise man once uttered that you ignore people at your own peril.