The story so far…NASA releases a set of data which they claim validates that they won the August 2017 Presidential election and on 26 Jan I am sent a 17 Mb PDF file which purports to be those results.
On first review, the detailed file looks superficially plausible, but the summary appears from two of two samples (Lamu and Tana River) to be a gross fake where someone has overtyped the real numbers (in the detailed attachment) with new numbers they made up, creating impossibly high turnouts which bear no relationship to their own source. In these counties, it seems NASA is trying to “rig” the polls in a crude and blatant way which can be disproved by anyone with a spreadsheet and a few hours, which if confirmed would surely be the end of whatever credibility NASA and Odinga had left.
I wondered initially if this file (which came without origin attribution) might be an incredibly clever ‘dirty tricks’ op by the Jubilee side, but then NASA released their results to the public, not as a PDF but as screenshots of that PDF, here.
From a few checks, the documents look the same. So I think NASA's credibility and claims can be assessed, and stand or fall, on the truthfulness of the PDF, which appears to be truly their work.
With another day to look at the file, I am confidential that the polling station file began as a genuine hack of the IEBC (which is of course illegal, something I won't consider further). Whilst anyone can fake anything with enough time and attention, this would be a huge project with 40,883 polling stations, and doing so by duplicating stations, or doubling numbers, would soon become visible because registered voters would be wrong or turnout too high. Looking at the PDF line by line for a particular county (Isiolo), I cross referenced polling stations and registered voters and I am satisfied with the tools at my disposal today that this was at some point A data set from the August 2017 elections.
It is not of course, the 2017 Presidential election results. As explained in many places including my column for The Elephant during June-September, the official results are never from the IEBC servers but from the Form 34As and Bs, paper records which should match the electronic results up to a point but which take precedence if they do not. There are lots of reasons why the electronic and paper systems might not match (without imputing foul play), including keying and transcription errors when typing numbers in to the EVID system, and polling stations where the electronic systems failed entirely or in part, and simple arithmetical errors by tired clerks counting up polling stations manually on the paper system. Another technical reason why the electronic system and the paper would not match is double entry of electronic data. Though it should not have been possible, the file NASA released has a few cases where the same results were posted twice for the same polling station (slightly differently). I don’t think they were included in the IEBC totals, but if you don't manually find and exclude them one would become quite confused..
However, I struggle to imagine a situation where the electronic transmission system shows a result in a polling station which has a much larger number of votes for a candidate than the final paper return, let alone that we see this for most of the polling stations in a County.
Unfortunately, turning to Isiolo, that is exactly what we find. My third deep dive into NASA’s results returned a completely different outcome. Again, I have (because I downloaded them painstakingly from the IEBC servers while they were online) the original Form 34 Bs for Isiolo North and Isiolo South. I expected - positing that the detailed polling stations file could be genuine - to see that NASA's electronic results would bear some relationship to the paper, but that was not the case.
This was NASA's summary for Isiolo: an 82% turnout (implausible) with a narrow victory for Raila by 3000 votes.
The IEBC official forms give an entry different result - a 5000 vote victory to Kenyatta over Raila with a lower 74% turnout. Again someone is cheating or faking their returns. Interestingly, one of the Official 34Bs is completely unsigned by agents, for a plausible reason, but certainly a warning sign.
On the other hand, there is an obvious flaw in the NASA polling station file too, which I missed on Friday. While the other candidates' names are written in a longer form, the name for the last two are shortened to "Uhuru" and "Raila" - see below
Someone has overtyped them before the PDF was created, which would also allow them to swap over the names, giving all Uhuru's votes to Raila and vice versa (though I don't think that happened, as Uhuru would then win Nyanza and Raila in Kiambu). But it means that this file too, even if it was once genuine, has been somehow played with.
In Isiolo's case, once you remove seven double entries for polling stations, which NASA had not spotted (which is why their turnouts were too high), NASA's summary and detail match approximately. So at least they didn't just make up the numbers completely as it seems they did in Tana River and Lamu:
REGD TOTAL UHURU RAILA
IEBC Form 34Bs 75,355 54370 26,746 18,931
NASA Summary Claim 73,358 61690 25,573 28,233
NASA Detail 75,338 TBD 24,326 26,824
Whether a Raila victory is plausible in Isiolo though, I very much doubt, because of the parliamentary and Gubernatorial results. In the governorship, pro-Uhuru candidates won 95% of the vote and the ODM candidate polled 2%. In Parliament, in Isiolo North pro-Jubilee candidates won 2/3 of the vote, ODM didn't even field a candidate and the NASA (Wiper) candidate got only 2%. In the tiny Isiolo South, the NASA ANC candidate got 21%, NASA Wiper 5% and pro-Jubilee candidates 60%. A Raila victory would be completely contrary to the alignment of the main candidates and the voting patterns in all the other contests.
But the differences between the electronic return (which may of course have been "rigged" by NASA) and the paper returns are still inexplicable. In Isiolo South, they are simply an entirely different data set. Even Kenyatta's number have been reduced and recorded for Dida in the form 34B:
I remember finding the strong performance of Dida odd in Isiolo South in August, as he isn't even from the area and polled virtually no votes elsewhere in the country. So, more to follow.
This is my working file for verification if anyone wishes to review.