Wednesday, November 17, 2010

Whatcom County at 86,949 Votes Counted

Certification Date:11/23/2010
Number of Precincts:120
Number of Registered Voters:116576
Total Ballots Cast:86949
Ballots Left to Count:0
Next Ballot Count On:Nov 23 2010 5:30PM
Last Tabulated:Nov 17 2010 5:07PM
Voter Turnout:74.59%

Vincent Buys (Prefers Republican Party)
Kelli Linville (Prefers Democratic Party)

30594 - 30436 =158 difference between them now.
30594 + 30436 = 61030 total votes
158/61030 = 0.0025888907094871374733737506144519 percentage difference

So does Linville qualify for a manual recount now or not?

"A machine recount will occur if the election results for the top two candidates are separated by more than 150 votes and less than 2,000 votes."
"A manual recount will occur if the difference between the top two candidates is less than one-quarter of one percent and the two are separated by 150 votes or fewer."

Looks like she is eight votes short (!!)

So here's my pitch:

How much is Kelli's seat worth to the Democrats? How much do they stand behind their candidate? A machine recount doesn't do much but check the machine's ability to duplicate the original result. A hand recount can find important errors in the Sequoia 400C optical counting. What kind of errors? Here the brief   "all you need to know from me"  guide to signature detection anomalies from someone whose job it was to test the validity of signature detection mechanisms on state of the art Intrusion Detection Systems....

There are always at least two type of signature detection errors made by automated systems:

(1) False Positives
(2) False Negatives

Let me explain the differences from a malware detection perspective and then apply the concepts to vote counting machines.

False Positives
I am testing whether a intrusion detection system responds to a particular piece of code/malware/network intrusion. So I create a signature for it, but I find my signature triggers not (just) on my malware, but on other behaviors, perhaps some that are benevolent.  This is a false positive: something that shouldn't have triggered the signature did, and I can see that result. False positives therefore almost always present us with the possibility of debugging and correction. I can simply rewrite and retest the signature until I get it right.
False Negatives
I am testing whether a intrusion detection system responds to a particular piece of code/malware/network intrusion. So I create a signature for it, but I find my signature won't trigger or triggers irregularly  on my malware. False negatives can present us with a conflict in production environments because there won't necessarily be any evidence to suggest the malware went undetected!

Thus, false positives are incorrect but ultimately solveable; false negatives provide no evidence of failure. False negatives are the scarier of the two. It is the false negative that kills you  in intrusion detection because you may never get logging data  to learn from bad input. Your client's network is ex-filtrated, his systems are compromised, and you and your client may never know what hit you or how.

Vote counting machines have a certain logic for the detection false positives. Mark both the yes and the no arrows for a particular vote. The machine should  detect this and spit out the ballot for remediation. But on a double sided ballot, 'bleed through' could  lead to a lack of detection that would be repeated on any machine count and so return the same faulty count. Consider the pictures of lined up arrows that I show in this post.  Someone marks hard a yes vote on one side whose arrow is directly aligned to a vote on the flip side of the ballot on an issue that the voter chose not to make a decision on.  So there is no 'misvote' mark to catch. Ever. Just a 'bleedthrough' that looks just like an intentional flip side ballot vote! Your automation never sees the problem...

Ultimately, the presences of potential false negatives on a double sided ballot argue most strongly for manual (hand) recounts. An obvious problem is that humans will make mistakes over the counting of 60,0000 ballots. But with enough manpower and checking, that problem is solvable. Another problem is the hand recount might not go your way. Another problem is that it will cost you a chunk of change.

But so does poor and inadequate electoral representation, so why not pay for a hand recount if you are within striking distance? . If the Democrats back Kelli, they should pay up for a manual hand recount at this point. Send observers. Lots of them. That's my two cents!

No comments: