The problem is only peripherally like a jigsaw puzzle. While it is possible to attack certain portions of the assembly problem using automated methods (namely breaking the chads into individually manipulatable pieces and - perhaps - suggesting pieces), the basic problem is serial in nature. In the latter puzzles such as #4 an #5, the primary issues are:
- the chads are offset with respect to one another
- the pages have been shredded slightly skewed
- the edges are poorly defined
- the top and bottom edges often overlap slightly due to the cut/rip angle that chops the strips into chads
- the shape of the chads have often been warped during the scanning process
But worst of all, given a particular chad, a human or computer must predict the appearance of the adjacent chad and then find it among thousands of possible candidates. Even after narrowing down the search by categorizing the chads into whatever groups seem useful, you often end up with multiple chads that will *exactly* fit in place. By that I mean that the writing appears - pixel for pixel - to continue onto the next chad. While one would think that human handwriting documents would be highly random, they aren't. We tend to use the same line angles when connecting cursive letters, crossing our t's or other writing gestures and this causes a high degree of commonality in the graphics at the magnified level of the chads at which we are working.
But wait, it gets worse - if you misplace a chad, you have actually created *two* errors in the document - the misplaced chad and the (now missing) chad where it *should* have gone. In a crowd sourced solution with many inexperienced eyes working part-time, it is my opinion that many of these types of errors will be introduced, preventing a solution. For this reason - and for this particular challenge - I think it will actually be won by some masochistic puzzler with some image-manipulation skills.