Here's another question: what a friend of the victim had testified that the victim had called sometime around the believed time of death and choked out, "Guede and Knox did it!"? If we're imagining that we don't know if there is a murder, then someone doing that isn't all that damnning, since hey, "it" could be any silly little thing? Yet it sounds pretty bad. Aha! Guede's name there associates something he did with Knox and it's agreed that he only did one thing of note in her apartment that day. OK, can we make a fact that would be more entangled? Yes: what if she said in that dying phone call, "Knox killed me"? Then aren't we saying that we only add the additional evidence of a roommate who isn't thought to be dead uttering that sentence? People say things like that all the time using "killed" as a metaphor for all sorts of innocent things. Yet it seems pretty damning. Can we really untangle the murder from the other evidence? I mean, the probability that the victim would at that moment use the word "killed" to mean something other than "stuck a knife in my throat" seems really low. And I think there is some similar entanglement with the other facts I've mentioned, too. The fact of the murder seems to make this technique faulty.
The OP would say that his framework is the only rational one.
The legal system doesn't work on Bayesian analysis. It has a probability system defined based on the meaning of English words... http://en.wikipedia.org/wiki/Burden_of_proof#Standards_for_d...
To me, I think this first line should read ".. since it's agreed that Guede was found guilty for his involvement in a murder that, due to extensive crime scene analysis, was clearly committed by multiple assailants". So I'm not sure why we must "start over, from scratch, as if the victim were still alive".
Furthermore, keep in mind that the evidence includes an initial written statement from Knox that implicates her in the murder.
False written statements are often signed in police stations.
Specifically parts of the article that discredit the body of evidence from the Judge's report (http://www.truejustice.org/ee/index.php?/tjmk/comments/under...) that supports a multiple-assailant theory?
And yes, the priors - like the probability that a 20 year old female college student with no history of violence or mental disturbance of any kind would commit a murder - are certainly very relevant. That is one of the least violent categorizations of humans you can create. (In fact a more accurate prior would be even more miniscule: the probability that such a person would lead an elaborate team sex murder, an act so rare there is no term for it.)
(However, there is some other evidence that is hard to deal with in this model, like the crime lab saying that there is Knox DNA on the bloody bra clasp of the victim. That evidence seems hopelessly entangled with the murder. How can this model be used to reason about it?)
Anyway, all I know of the case is what other commenters here and the OP on lesswrong.com wrote, so my facts are suspect. I've just tried to explain what the OP is saying. Did I succeed?