Skip to content

Popular captchas that should be improved

edited April 2013 in Need Help
Real success rate is not actually totally real. It is just software was able to generate a guess
These data collected from over 10 million unique urls - randomized - small portion of it processed though
I have only written popular ones we should not spend time on not popular ones

Solve media : 3 / 369 - while success rate is defined as 40% - real success rate is 3%
Mollom : 399 / 2850 - success rate is 11% - real success rate : 14%
Ard Guestbook (color) : 0 / 209 - success rate is 100% - real success rate : 0%
Burning Board (type u186) : 6 / 2786 - success rate is 100% - real success rate : 0%

XMB (white grey) : 15 / 117 - success rate is 15% - real success rate : 13%
YaBB : 14 / 94 - success rate is 31% - real success rate : 15%
kohany : 10 / 66 - success rate is 23% - real success rate : 15%
drupal (8 chars): 18 / 122 - success rate is 13% - real success rate : 14%


Comments

  • OzzOzz
    edited April 2013
    don't confuse the definitions. not every answer that was given was a correct one. 

    take mollom as an example. there were 399 of 2850 captchas CB gave an answer for. but how many of those 399 were correctly solved? i guess not that many as mollom is hard to solve. 

    the percentages you see are all based on a sample size where the captcha was tested on with the SDK. to get a "real" number you need a good sample size with correct answers to test in the SDK. the larger the database that Sven has for those types of captchas, the better the real number could be calculated.

    you should also keep in mind that some of your examples lack in sample size. with 100 captchas that were sent to you in that particular time frame you can't know if you have good or bad luck with them. within this 100 captchas there could be 50 correct solves one day and only 10 the other day.

    however, the 100%ers should solve it with 100%. i don't know what is wrong with that definition. you should save those captchas to file, answer them correctly and send those samples to Sven to speed things up.
  • edited April 2013
    @Ozz i know. i just mentioned these are popular ones among all and should be improved

    and with those %0 definitely they are broken
    probably those platforms updated their catpcha

    and they are also common ones
  • either they are broken or are false detected. this could happen with activated "try to solve unknown captchas" for example.
  • well that option is not activated :)
  • LeeGLeeG Eating your first bourne

    I was watching the captchas yesterday

    For some reason Freglobes gets picked up as Jambook Guestbook

    Freeglobes is a 100% success rate. When its seen as Jambook, its 0%

  • @Leeg that seems very logical. CB need an update :)
  • SvenSven www.GSA-Online.de
    Just tested freeglobes...works fine on my end. @LeeG and all who have problems with a particular captcha fail detection...I need the captcha itself to debug.
  • Hi some trick with reCaptcha ... if you didn't know maybe useful...

    i made some screenshots of reCaptcha and the typed in answers that show correct

    you can test at http://www.google.com/recaptcha/learnmore


    image

    so as you see only the "cursive" part of the captcha has to fit .. for the other you can enter whatever you want
  • LeeGLeeG Eating your first bourne

    Here you go Sven.

    I used these to set up Jambook as freeglobes

    https://www.dropbox.com/sh/wr1cgt06ljrh8yt/_PmqAC2R9H

  • @KayKay when I tried the same thing some time ago, I found out that if you use http://www.google.com/recaptcha/learnmore the behavior is basically as you describe, but when you do the same with a real recaptcha in any signup form, the behavior is more strict, at least I had that problem.
  • SvenSven www.GSA-Online.de
    thanks for the captchas...working on that now...
  • SvenSven www.GSA-Online.de
    @LeeG, loading those freeglobes works perfectly here. Did you change anything on your captcha types from the default once?
  • OzzOzz
    edited April 2013
    i noticed this a moment ago:
    image

    this are the captchas that were used:

    when i put them in the SDK and test than some of them work or give an answer at least.
    the log says "Article - UCenter" and should be "Social Network - UCenter" but i doubt that this is the issue??!
  • LeeGLeeG Eating your first bourne

    In all honesty Sven, I have no idea on that. I have a work around in place anyway

    It might even be a residue from when I broke the captcha sdk system in the beta days :D

  • SvenSven www.GSA-Online.de
    @Ozz seems you used so called "optimized" engine files ;) "Article - Ucenter" is the name given by SER as the engine file is called like that. CB however only knows UCenter.
  • LeeGLeeG Eating your first bourne

    So have renamed engine files can cause the problem. Not that any of us would be doing that by any means.

    I blame Ozz, its all his fault

    Exit stage left at high speed

  • SvenSven www.GSA-Online.de
    edited April 2013
    well a renamed engine file could lower the success of detecting the correct captcha type, but thats all.
  • OzzOzz
    edited April 2013
    k, good to know :). is it possible to use wildcards for detecting like "*UCenter"?
  • SvenSven www.GSA-Online.de
    edited April 2013

    hmm not for now...but i can add that for next version

    Just had a look in the source, you can use that already.

  • OzzOzz
    edited April 2013
    k, implemented by accident then :)
    could become handy especially for Web20 kind of platforms where you never know how other tools named the site like "www.example.tld" or "example.tld" or just "example". now "*example*" should work for everything.
Sign In or Register to comment.