It turns out the Obama campaign wasn’t the only group to put together a data mining project for the presidential campaign. The Romney campaign has a project they titled Project Orca. Project Orca came at the Election Day problem from a different direction than Narwhal did. In Orca, the engine was designed to track exit polls, entry polls, and monitor the voter registries to see who had voted.
But Orca ran into a bit of a problem. Well, a huge problem if the news is correct. Not only did Orca not get properly tested before Election Day, but power outages and mismanaged communications killed the project before it got off the ground. But that apparently isn’t the worst of it. According to one volunteer, the miscommunication extended to giving volunteers the wrong PIN access code (in multiple states)–and issue that didn’t get fixed until 6 p.m. (well after voting had closed in many states), telling volunteers they were responsible for their own materials, and forgetting to tell volunteers to bring in their poll credentials so they COULD get to work.
In the original article I read (which I can’t find right now), it stated that not only was Orca not tested, but that a majority of high ranking campaign staffers didn’t even know the app existed. The brains behind the project decided that either they couldn’t be trusted to keep their mouths shut or that if they were told about it, the Obama campaign might steal their thunder.
Well, that’s one thing they didn’t have to worry about. The Obama campaign already had their secret project, which all their staffers knew about, which combined multiple databases, and self-corrected during the course of the campaign. In fact, if anything it was the need for secrecy and compartmentalization that killed Project Orca.
Data integrity, secrecy, and proprietary code are three of the biggest issues DBAs have to deal with. Most first world countries have a myriad of privacy laws to contend with that make protecting the data a number one issue. But the one thing DBAs can’t do, the one thing that would definitely keep the data private, is to unplug the server. We have to uphold the privacy laws, protect our employers’ proprietary practices, and keep the data safe all while serving it up to our customers. In this, the Romney campaign failed spectacularly.
In SQL Server, table structures and queries can easily be hidden from the end users by implementing security based on Schemas, using views to obscure tables, and calling the views or stored procedures on the client end without exposing real table / column names and how those tables are “keyed” together (foreign keys / relational references). It’s not just SQL Server that has these tools. I believe Oracle has them, not to mention the plethora of third party tools used to massage the information into data and then slice, dice, and present the finished reports on someone’s little screen.
If the techheads behind Project Orca were so paranoid, than this is how they should have done it. They also should have tested the software months in advance of election day, had the poll watcher volunteers go through dry runs using post-debate polls (which would have put a nice test load on the servers), and made sure that all needed power was accounted for and had backups. If they’d bothered to do some Quality Assurance testing, if they had bothered to have Orca up and running during the course of the campaign, then perhaps the Romney staffers wouldn’t have been so shocked when things fell apart at the last moment.
According to the volunteer, Project Orca was showing a lot of states in red and pink, even the states that were really blue. If this is the case, it’s hardly a wonder Governor Romney and Rep. Ryan got a little shellshock when the networks called the election for President Obama. If my untested data mining application was lying to me, I’d be shellshocked too.
I’d also be fired and out of a job. In the real world, businesses would lose serious money on a fail-event like that. Lots of bad data that bleeds millions of dollars equals out of work programmers, DBAs, and sometimes even their managers.
That’s what really happens when data mining goes wrong.
