The Risks Digest

The RISKS Digest

Forum on Risks to the Public in Computers and Related Systems

ACM Committee on Computers and Public Policy, Peter G. Neumann, moderator

Volume 9 Issue 40

Friday 10 November 1989


o "Computer Error" in Durham N.C. election results
J. Dean Brock
Ronnie W. Smith
John A. Board
o Glitch in Virginia election totals
Paul Ammann
o Rome: Operator error causes publication of wrong election results
Lorenzo Strigini
o Delayed Stock Exchange Opening
Brian M. Clapper
o Electronic Warfare Systems not working--Congress
o Computer used to find scoflaws in Boston
Peter Jones
o Computer errors and computer risks
Randall Davis
o Equinox program on Airbus
Lindsay F. Marshall
o Info on RISKS (comp.risks)

"Computer Error" in Durham N.C. election results

J. Dean Brock <>
Thu, 9 Nov 89 14:00:12 EST
The headline on the November 9, 1989, Durham Morning Herald is
    Computer Twists Election Results
According to the article a "computer error" caused eight precincts
to be counted twice.  The correction actually changed the result
of one city council race twelve hours after it was assumed settled.

It's difficult to determine the nature of this computer error
from the newspaper article.  Another front-page article entitled:
    "Haywire Machine Counted Precinct Vote Totals Twice"
quotes Jo Overman, the chairman of the County Board of elections, as saying:

    One terminal used Tuesday apparently counted twice
    each precinct entered into it....
    What was called in was correct, the computer just added it twice....
    It was not added by an operator, it was a glitch in the program.

Ms. Overman, also added that the "errant terminal was an extra unit put on
election duty as part of a last-minute effort to process returns faster."

Interestingly, the precinct-by-precinct breakdown given to the media was
correct, even though they did not match the totals.  The mistake was
discovered in a later hand check of the results by the Board of Elections.
Apparently, no one else bothered to check the totals.

The director of the county's Management Information Services department,
which would be responsible for any programming errors, was instructed
by the elections supervisor not to say anything about the election.

"Computer Error" in Durham N.C. election results

Ronnie W. Smith <>
Fri, 10 Nov 89 08:55:29 EST
The only information I have to add is that the local TV media kept referring to
it as a "computer error" without ever mentioning that the original source of
the error was a person.  The newspaper never explicitly made this link, but at
least mentioned it was a programming error.  Interestingly enough, the man who
became the winner after he had been declared the loser did refer to it as a
"human error".  The number of votes that had been double added was slightly
more than 6000.

"Computer Error" in Durham N.C. election results

John A. Board <>
Fri, 10 Nov 89 12:22:16 est
[...] I find it most fascinating and troubling that it took over a day for
anyone to notice that the correctly reported precinct votes duly tabulated in
the paper the morning after the election did not add up to the numbers reported
as totals at the bottom of the columns, and the errors were not small - the
Mayor's race vote, for example, had been reported as 19,381 to 17,118 when in
fact the real totals of the votes as listed were 16,136 to 13,356!  To the
credit of the elections board, the errors were apparently found during manual
verification of the automatically reported "unofficial" results.

           [With a "Duke" as Governor of both Massachusetts and California,
           I wonder if any Duke Univ. folks were governing this election?  PGN]

Glitch in Virginia election totals

Paul Ammann, George Mason University <>
Thu, 9 Nov 89 14:25:22 -0500
In the Nov. 7 Virginia gubernatorial race, Doug Wilder (D) appears to have
defeated Marshall Coleman (R) in a close race.  Currently out of a total of 1.7
million votes, AP reports a difference of 5,533 votes and UPI reports a 7,755
vote gap.  The Post article referenced below discusses the reasons for the
discrepancies and the mechanism for official vote tallies.  Buried within the
article was the following gem: (Washington Post, Thursday, Nov 9, 1989, pp.
A37, A40.)

  Vote Counting Methods, Race Factor in Polls Leave Plenty of Room For Error
  Disparities Remain in Va. Governor's Race Tallies
  By Stephen C. Fehr, Washington Post Staff Writer

  [discussion of discrepancies between AP and UPI vote tallies]
  For an hour on Tuesday, [AP's director of planning Evans] Witt said,
  a computer glitch caused some of Wilder's votes in predominantly black
  precincts to be counted twice; the error was fixed and the vote total
  was adjusted.
  AP's Witt said that "there's almost always a variation between the
  official and the unofficial count," but said he could not think of an
  instance in which the results of an election had been reversed because
  of a mistake by the wire services.

There were many surprises and mistakes in the projections and reports of
election results; the type of problem cited above is minor, but, left
undiscovered, potentially quite serious.  The decision to call for a recount,
as well who bears the cost of a recount, depends upon the closeness of the
election (according to the official tally, of course, which is due on the
fourth Monday of November).

Rome: Operator error causes publication of wrong election results

Fri, 10 Nov 89 09:37:51 SET
On October 29-30, elections were held in Rome for a new city administration.
Unofficial results published at first gave an important victory to the Christian
Democrats, but at the end of the tallying this victory almost vanished. The
publication of wrong results was attributed to a data-entry operator error.
Since then, the political parties have been exchanging accusations of
intentionally manipulating the data for political advantage (the supposed
advantage would be a short-term boost in popularity for the Christian Democrats,
or casting suspicions on the Christian Democrats, for the Communist). To set
things in context: besides deciding who will manage the capital city of Italy,
these elections were regarded as an important indicator for national policy, and
the major parties had put much effort in a combative, venomous campaign.

Now the details. (Disclaimer: this is my interpretation, checked with a few
colleagues, of very imprecise press and radio reports.  I'll look for more
precise reports, and send in corrections if I can)

Voting and vote counting are by hand, with paper ballots.  After the count
started, and as partial results were transmitted from the individual "electoral
sections", an EDP center of the City of Rome added them to obtain partial
accrued results (without official value) and transmitted them to the press,
radio and TV.

Very soon in this process, the published results showed a marked gain for the
Christian Democrats. Later, it turned out that a few tens of thousands of extra
votes had been erroneously given to them.  The error became evident because the
sum of the votes was greater than the number of voters. In an interview, the
director of the EDP center stated that he had received from the computer program
warnings about the discrepancy, but had ordered the publication of results to
continue, assuming the problem was temporary and it would disappear later on.

Two days ago, the operator was found that allegedly caused the problem. He had
to type in a screenful of data, send them to the computer and wait for it to
clear the screen and prompt for new data (or to unlock the keyboard?). He found
that pressing a certain combination of keys allowed him to clear the screen and
restart input sooner, so speeding up his work. But by this trick he sent wrong
data ("this affected the votes for 4 parties, and in particular the
number of votes for the Christian Democrats -line 18 on the screen - was
substituted with the number of the electoral section").  The program would
complain about receiving inconsistent data, but give him an override option,
which he used.

Now my comments. Funny: everybody is complaining about evil intentions (of which
there's no proof), not about incompetence. From the news stories, some
technical/organizational flaws are evident:

- the input routines checked for transmission overruns, or the application
program ran consistency checks on each individual transaction (the entering of
the results from a given number of ballots) but allowed the operator to override
them (there was a log of the override requests, though: all inputs were logged
to tape; but the log of part of the session was lost because tapes were scarce,
and some were used twice).

- the director of the EDP center ignored the warnings (it is unclear whether
these were from a global auditing of the data base or were the same error
messages sent to the operator) about inconsistent data.

But, most important: in their greed for early results, both the press and the
politicians trusted a non-trustworthy system.  It appears that the only checks
applied to this unofficial counting procedure were the consistency checks
mentioned. If one were to bribe the operators to shift votes _consistently_ from
one party to another, this could go undetected until the official tally was
available, several days later. The vulnerability so created is great: news
reports of, say, an 80 % victory of the Communist Party would certainly hit the
Stock Exchange hard; the resulting allegations of fraud would cause a political
earthquake (in the '50s, they might well cause a civil war).  As things are, the
effect on the public appears quite serious: according to an opinion poll, some
30 % of the voters interviewed said that, if the election were held again after
the news of the mix-up was known, they would refuse to vote.

Lorenzo Strigini
Istituto di Elaborazione dell'Informazione, Pisa, Italy , strigini@icnucevm.bitnet
IEI-CNR   Via Santa Maria 46   I-56100 Pisa   ITALY

   [Regarding greed for early results, it was interesting to note that the
   advance polls in the New York City mayoral race were off by roughly
   11%, and the exit polls were off by 10%.  PGN]

Delayed Stock Exchange Opening

Brian M. Clapper <bmc@SEI.CMU.EDU>
Fri, 10 Nov 89 11:53:05 EST
I received the following information from a friend of mine, William Power,
who works as a reporter for the Wall Street Journal.

  The New York Stock Exchange (NYSE) and the American Stock Exchange (AMEX)
  opened for trading approximately one hour late this morning (November 10) due
  to an inability to receive information from or transmit information to the
  Securities Industry Automation Corporation (SIAC), the jointly owned computer
  processing subsidiary of the two exchanges.  SIAC suffered equipment damage
  due to a fire in its building at 55 Water Street in lower Manhattan.  The
  fire apparently damaged equipment in a basement electrical vault, resulting
  in power outages to some areas of the building.

  The initial fire alarm was posted at 8 am; the NYSE and the AMEX officially
  opened for trading at 10:30 am, one hour later than usual.  The delayed
  opening resulted in a "domino effect," including the partial shutdown of the
  Chicago Mercantile Exchange.

Brian Clapper, Software Engineering Institute, Pittsburgh, PA 15213

Electronic Warfare Systems not working--Congress

8 Nov 89 14:19:36 GMT
The Nov. 7 issue of the *Washington Post* carries a front page article on the
failure of long-term EW development projects to deliver on their goals and to
adequately counter 20-year old threat techniques.

The article describes a Congressional study, to be published soon, that looked
at the B-2 bomber, and a service-wide EW system, now in its thirteenth year of
development.  Of particular interest is the criticism of the test methods,
described as not keeping up with teh technology to be tested.

Although the article doesn't mention software specifically, the B2 software has
been a significant issue.

My own experience in EW systems is that black projects seem to engender the
attitude that since the project is not as visible, we can get away with less
formal control and more ad hoc technical approaches.

Disclaimer:  the truer it is, the stronger the denial.

Computer used to find scoflaws in Boston

Peter Jones <MAINT@UQAM.bitnet>
Tue, 7 Nov 89 18:05:59 EST
On Sun, 5 Nov 89 13:14:43 EST, "Barry C. Nelson" <>,
in RISKS  Volume 9 : Issue 39 said:
>When five out of six hits are human errors, imagine the complaints!

It goes to show the importance of considering the total effect of a system
change, not just the project at hand. It was a serious design error to assume
that licence numbers, even if they could be read accurately from a TV camera,
could be used to positively identify wanted vehicles, if the database that
indicates which numbers are "hot" is unreliable.

Peter Jones     MAINT@UQAM     (514)-987-3542
"Life's too short to try and fill up every minute of it" :-)

Computer errors and computer risks (e.g., RISKS-9.39)

Randall Davis <>
Thu, 9 Nov 89 15:40:17 est
Numerous stories have been reported on this list under the title "computer
error" and "computer risk," that seem to me to have nothing essential to do
with computers, and a great deal to do with very different issues.

Consider this story, for instance, from 9.39:

>Subject:      new computer risk: child abuse data base proposed
>     According to a news release heard a day or two ago, MI is now considering
>legislation permitting local communities to establish and maintain data bases
>of "suspected" child abusers, or those meeting another of the nebulous
>"profiles" used to identify all sorts of persons and ethnic groups in our
>society. Aside from permitting hearsay from neighbors, teachers, co-workers,
>associates and assorted third parties to be entered and disseminated,
>the framers of this legislation are also attempting to gain back-door access
>to medical records. One profile criteria disclosed for "identifying" child
>abusers is use of multiple doctors/hospitals by the same family....
>Obviously, the privacy considerations and potential for misuse and/or
>malicious use, such as slanderous reports by neighbors against an unpopular
>neighborhood resident, inherent in this legislation are enormous.

If this is essentially a computer risk, there is an easy solution: get rid of
the computer and we get rid of the risk.  Modify the legislation to require
that all database records must be kept manually on paper.  If this story is
really about computer risk, then all the problems noted above will disappear
when the source of risk is removed.  But do they?

Of course not.  Because the problems are privacy, vague definitions, hearsay,
backdoor entry, and interference in our lives.  The technology used to
accomplish those things is of some consequence (typically it changes the
economics), but it is not of the essence.  The real problems existed long
before this particular technology and are largely independent of it.

It matters how we describe these things because descriptions implicitly set
the agenda for discussion.  To call it a "computer risk" is to set an agenda
for discussing computers.  This is particularly misguided when the questions
that ought to be asked are: Should we collect such information at all?  HOW we
collect and store it will eventually matter, but the first and fundamental
question is, shall we do it at all?  What rights to privacy do we have?  What
modifications are we willing to make to those rights in pursuit of other,
clearing conflicting goals in society?

In Mass., for example, (and perhaps elsewhere) doctors are required to report
to a state agency evidence of child abuse (not just obvious cases, evidence).
This is clearly a risky violation of the privacy of the doctor/patient
relation, one that includes most of the problems noted above.  The risks are
reduced here because the information required is a professional opinion based
on physical evidence.  In this case it is a risk we accept, presumably because
we believe the tradeoff is worth it.  And *that's* what the discussion ought
to be about: the risks and benefits of what we are doing, not what technology
is used.  The risks and benefits are often magnified by the technology, but
the essential question is the risks/benefits of various sorts of privacy and
the character of the information collected, not the technology that happens to
be employed.

Of all groups, this list ought to get this right.  Let me thereby enter a plea
to use the term "computer risk" and "computer error" with considerable
technical discretion.  I suggest the simple test above: Ask, can the identical
problem can arise in the absence of computers?

In some cases the answer is no (eg, instant, large-scale access to data from
arbitrary distances), and these are essential, computer-related risks.

But if the same problem can arise, it is quite likely the technology is
fundamentally irrelevant and that the risk involves something else.  In that
circumstance ask what the problem is normally called and use that name.  The
story above, for example, is about risks to privacy, the dangers of using
inaccurate information from questionable sources, and requiring people to
report one another's activities.  Removing the computer from the picture does
not change those problems in any fundamental way.  And the problems are
serious enough that they ought to be debated on their own terms, without
muddying the waters with technology.

Equinox program on Airbus

"Lindsay F. Marshall" <>
Thu, 9 Nov 89 12:50:03 BST
I managed to get round to watching the program last night and found it very
interesting. The program was very smooth except for one sound glitch - which
occured right in the middle of the word "reliability" when the narrator was
discussing the multiple processor architecture...

Lindsay Marshall, Computing Laboratory, The University, Newcastle upon Tyne, UK

Please report problems with the web pages to the maintainer