Monday, May 24, 2010

The Comelec pulled off a fast count alright, but is it accurate?

(Roberto Verzola has a background in engineering and economics and a passion for social and environmental issues. He is recognized by the IT industry as an Internet pioneer in the Philippines and works with NGOs on technical issues. He is currently a lecturer at the Institute of Mathematics of the University of the Philippines and a convenor of Halalang Marangal (HALAL).

By Roberto Verzola


The May 10 vote count reached millions within a few hours, and was 90% complete within a week. That’s supposed to be impressive. Stunning, even.

But everytime I ask, “so, how accurate are the counting machines?” I get a blank stare, including a surprising “I don’t know” from Mr. Gene Gregorio, Smartmatic spokesman. He gave this
answer before journalists at the May 8 Kapihan sa Sulo forum, two days before the elections.

I had also asked Commissioner Gregorio Larrazabal on April 30, when he met a team of IT experts convened by candidate Joey de Venecia III. His reply was also “I don’t know”.

Nobody, it seems, knows. Not the political parties, not PPCRV and other election watchdogs, not the media.

So, we spent around P11 billion and held a nationwide electoral exercise to determine the future of our nation, without knowing if the machines that counted the votes are accurate or not.

By the way, once we know the accuracy, we can get the error rate. A machine that is 95% accurate has an error rate of 5%. One that is 99% accurate has an error rate of 1%. Comelec specifications called for 99.995% accuracy. This means an error rate of .005% or lower no more than one error in 20,000 marks.

We need such low error rates for very close contests. An accuracy of 99.9% (0.1% error rate) may seem good enough. But such machines cannot resolve contests where the apparent winning margin is less than 0.1%. That margin may be simply due to machine error. In 2007, for example, Zubiri supposedly got .07% more votes than Pimentel, which won Zubiri the 12th and last slot for senators. Machines with 0.1% error rates or higher would have been useless for resolving this contest.

The real test of the 2010 automated election is the very close contests, many of which are still unresolved. For such contests, we need low error rates.

We had six chances to know the PCOS error rates. In five, the Comelec either kept the results from us or otherwise undermined our chance to know. In one, we got a really good idea of the
PCOS error rates. Let us go through each of the six, one by one:

1. SysTest Labs system audit and source code review (half a year before the elections). We paid SysTest some P72 million (1% of total project cost) to conduct a system audit and source code
review of the system. One of the things they should have measured was the error rate of several representative machines. When the Comelec released part of the Systest reports on May 1,
I immediately looked for the machine error rate. I didn’t find it. Either SysTest did not measure it – a major omission – or the Comelec kept the results confidential.

2. Comelec acceptance tests (several months before the elections). As PCOS machines came in, the Comelec should have tested these for error rates, among other things. This is simple
due diligence before accepting an expensive delivery from a vendor. Any machine with an error rate higher than .005% should have been returned to Smartmatic for calibration or replacement.
Accepted machines should have the test results attached to them, accessible to any inspector or stakeholder. The Comelec, if it did these tests at all, have kept the results confidential.

3. Final testing and sealing (FTS) (three days before election day). In some areas, the FTS was done earlier and to everyone’s dismay, the machines made grievous errors! The results were so
bad that the Comelec hastily ordered all election inspectors to stop further testing of the machines. For once, we had a good idea of the machines' accuracy, or lack of it.

4. Second FTS. That fiasco triggered a last-minute mad rush to recall, import, reconfigure, redistribute and reinstall new memory cards in time for May 10. In the chaos, security and
chain of custody procedures must have been ignored or bypassed. Were all new memory cards properly configured? Were all properly delivered and installed? Were all machines properly tested? Did all machines pass the test? I've heard this story several times: “We were told that the testing will be done Sunday afternoon; when we went Sunday, they told us it was done
Saturday.” Thus, when we held the elections, we did not know which machines were accurate, and which have remained grossly inaccurate, as we saw in the first FTS.

5. Voter verification of correct scanning of voters' choices (on election day). This feature, which is built into the machine and is required by law, would have displayed on the screen the names
of candidates corresponding to the ovals which the voter marked, a confirmation that the machine accurately registered the voter’s choices. This feature was disabled by the Comelec. Thus, if the machine was registering candidates other than their choices, the voters would never know.

6. Random manual audit (after the elections). Unfortunately, this audit has lost credibility. First, they announced the precincts to be audited noontime of election day. Forewarned which
machines would be audited, the cheats would have ordered their field operators to stay away from these precincts. A normal audit should finish in half a day — one day at most. Yet, three
days after the elections, no results had been announced. Subsequent results that came in were not made public. The Comelec simply made general public statements that “no discrepancies were found”. Results were delayed a few days, they explained, because ballot boxes were already sent to the municipal treasurer’s office where they had to be retrieved. That's enough time and opportunity to substitute ballot boxes.

Despite these, the Comelec and local election authorities have already proclaimed winners. They did not even await the random audit results, as if they already knew that the audit would
simply confirm the results.

The public is being told, it seems: “We’ve already pulled off a fast count, now you want us to be accurate too?”

No comments: