More than a year ago, a reporter named Craig Unger asked me a brilliant question about the 2004 presidential election in Ohio. Referring to reports that the counting of votes was rerouted through SMARTech computers in Chattanooga, Tennessee, at precisely 11:14 PM EST, and the possibility that the vote count was altered thereby, he asked me about the forensic evidence we had photographed or otherwise uncovered in eighteen counties in Ohio. He asked me to look at the timeline. Did the counties in which we found evidence of fraud report their results before 11:14 PM, or after? And, if after, was the evidence consistent with an after-the-fact effort to get the election records to match an altered vote count?

Unger forwarded to me a list from Bob Fitrakis. It stated, for 33 of 88 counties in Ohio, the very minute that the county reported its results on Election Night. Eleven counties are known to have reported their results by 11:14 PM, and I found no evidence of fraud in any of them. Twenty-five counties, including three not on Fitrakis' list, are known to have reported their results after 11:14 PM, and there is evidence of fraud in thirteen of them – Butler, Clermont, Clark, Cuyahoga, Delaware, Geauga, Hamilton, Lorain, Mahoning, Miami, Montgomery, Van Wert, and Warren.

The unanswered question is whether the forensic evidence suggests a connection. Had the election records in Ohio been altered in order to match a vote count altered in Tennessee? This report will examine all thirteen of the above-named counties in order to answer that question.

In Butler County, where the Board of Elections had always done its own programming, an ES&S technician showed up on Election Day and reprogrammed all of their tabulators. When we examined the ballots for eleven precincts, we found proof of ballot tampering in almost all of them. In three precincts we found long runs of consecutive ballots for Bush – 212 in Monroe City 4CA, 50 in Liberty Township 4DF, and 41 in Ross Township 4JB. In four precincts – St. Clair Township 4KA, Liberty Township 4DA, Liberty Township 4DI, and West Chester Township 5LE, more ballots exist and were counted than the total number issued to the voters. And in two precincts – Liberty Township 4DO and West Chester Township 4LM – the ballots did not match the tabulator count. Butler County was the first to agree to my request to examine the ballots. The Board of Elections was completely forthcoming with all election records. The circumstances are fully consistent with an outside operation. As it happens, Butler County gave to George W. Bush a 53,629-vote margin of victory, his largest of all the counties in Ohio.

Warren County was strictly a local operation. On Election Night, under the guise of a homeland security alert planned a week in advance by county officials, the county buildings were locked down, and there were no public observers of the vote count. Ballots punched for Kerry were shifted to precincts where, due to “ballot rotation” (see below), they would be counted for Bush, and only then were the precinct numbers printed on back sides of the ballots. Because the ballot tampering occurred on Election Night, before the ballots were run through the tabulators, the ballots match the tabulator count. It was almost the perfect crime, and any outside interference would have ruined it. But so many ballots were shifted that the results for president were highly unlikely when compared with other races; and the provisional ballots revealed how the crime was carried out. By law, all ballots issued at the polls, including provisional ballots, must have the county and precinct already printed on the back side. Regular ballots go directly into the ballot box. But each provisional ballot is placed in a separate sealed envelope signed by the voter, which is not examined until eleven days later. If and when it is determined that the vote should be counted, the envelope is opened. There is no opportunity to tamper with provisional ballots on Election Night. Thus, the provisional ballots with nothing printed on the back side, which we photographed, were the true examples of what the ballots looked like when they were issued at the polls.

Clermont County was at least in part a local operation. The infamous “stickered ballots,” on which little white stickers were placed over the marks for Kerry before the ballots were run through the scanners, were obviously the product of the Board of Elections. We audited thirteen large precincts, and we found many more ballots in existence and counted than the total number issued to the voters. Therefore, some of the ballots are fake. Filling out fake ballots could easily have been a local operation. But the ballots we photographed in these precincts do not quite match the official count, which suggests an imperfect after-the-fact attempt to doctor the evidence, to get the ballots to match an inflated vote count coming out of Chattanooga. At the partial “recount” staged in December 2004, precincts amounting to 3% of the ballots cast were counted by hand, in the presence of public observers. The precincts were supposed to be selected at random, in full public view. We have documented proof – an official memo, issued by the Clermont County Board of Elections six days before the “recount” – showing that the precincts to be hand-counted, the smallest in the county, had already been designated. Moreover, all absentee ballots for those precincts had been marked by hand, by Board of Elections officials, no later than Election Day, as a warning to leave the ballots for those precincts unaltered and able to withstand public scrutiny.

In Clark, Hamilton, and Montgomery counties, due to time constraints, we only examined uncounted ballots – that is, ballots cast with no vote recorded for president. These would be “undervotes” (ballots with no choice for president), and “overvotes” (ballots with two or more choices for president). In these urban counties (and also Cuyahoga, Lorain, Richland, Stark, Summit, and Trumbull counties), there were anomalously high percentages of uncounted ballots in many of the most heavily Democratic precincts. We were never allowed to look at the ballots for Lorain County; no other county successfully refused our request to do so. We audited all the uncounted ballots from 145 precincts in the other eight counties. Nearly two-thirds of them (2321 of 3512) contained multiple punches for president – usually a punch for Kerry, and for one or more third-party candidates – thus resulting in an “overvote.” Ballots with multiple punches were most common in precincts where Bush and Kerry were listed first, and third-party candidates afterward. Voters would punch their choice for president and move on to the next office, rarely noticing that their ballots had already been punched for another presidential candidate. The Boards of Elections in each county are independent entities; therefore, the fact that this happened in eight (or nine) different counties shows coordination coming from an outside source. The ballots, as punched, almost always matched the official count, which suggests that the ballots were punched in advance – most likely at the factory, before being shipped to the Boards of Elections – and that this method of vote rigging had nothing to do with SMARTech. But in seven of the eight counties, the exception being Cuyahoga (see below), we only looked at undervotes and overvotes, wanting to know why so many ballots were not counted. We do not know if the vote totals for the candidates were otherwise correct.

In Mahoning County, 20 to 30 touch screen voting machines “needed to be recalibrated” because votes for Kerry were flipping to Bush. The most revealing precinct was Youngstown 2E, where two voters selected Kerry and Bush's name came up, one voter had to scroll through five times before his votes were finally recorded, and three voters said that the presidential option never appeared at all while they were trying to vote, and that this happened on two different machines. And yet there were no “undervotes” for president in this precinct. Every voter who touched a screen had a vote counted for a presidential candidate. The simplest explanation is that the touch screen machines were programmed to default to a candidate of the vote riggers' choosing, the one whose name sometimes appeared against the will of the voters – George W. Bush. But it is also possible that “undervotes” were changed to votes for Bush, after the fact, by SMARTech. Because there were no paper ballots issued at the polls, it was not possible to audit the vote count. We have no idea how many votes were flipped from Kerry to Bush, or whether other electronic vote rigging methods were also utilized.

In Geauga County it was population data that told the tale. It is possible for the number of voters on the rolls to approach or even exceed the total voting age population. This is mainly because voters die, or move away, which is why voter rolls need to be “purged” from time to time. Also, voters might register more than once, just in case their names have been wrongfully purged. (It is not illegal to register more than once; it is only illegal to vote more than once). In Ohio, statewide, the number of registered voters was 69.6% of the total population (7,972,826 of 11,459,000), and this number was somewhat inflated because not every county had purged its voter rolls. But in Geauga County, in six entire townships, the number of registered voters ranged from 75% to 80% of the total population, even though the voter rolls had twice been purged since the 2000 election. In Hunting Valley Village, the number of registered voters equaled 95.63% of the total population (679 of 710). If all but 31 persons in Hunting Valley Village were registered to vote in 2004 and, according to the census, there were 159 children living in Hunting Valley Village in 2000, there must have been at least 128 children who died, or moved away, or ran away from home, or were kidnapped, or came of age and registered to vote. Either there were many missing children, or many bogus names upon the voter rolls, or numbers without names to back them up. These inflated numbers of registered voters were reported on or before Election Night, which would have given SMARTech an opportunity to inflate the vote count as much as they dared to do.

In Delaware County, official voter turnout was above 80% countywide, was above 85% in 27 of 123 precincts, and reached 91% in two precincts. (Statewide, the official voter turnout was 71.77%). According to official voter registration data obtained from the Ohio Secretary of State and the Delaware County Board of Elections, the number of registered voters increased from 85,925 to 100,676 between March 2, 2004 and November 2, 2004, despite the purging of 7,917 names on April 7, 2004. In response to a public records request, the Board of Elections was able to provide the names of only 16,435 persons who registered to vote during this time period – 6,233 short of the 22,668 names needed to verify the numbers. In addition, there were 7,184 “inactive voters” whose names were not purged until after the election. This would have opened a window of opportunity for SMARTech to inflate the vote count. Delaware County was the only county that refused our request to photograph or photocopy the voter signature books. But when we audited the ballots for Genoa Township, Precinct I, we found 359 consecutive ballots punched for George W. Bush.

In Van Wert County, in 2004, John Kerry received fewer votes than Al Gore had received in 2000, even though there was a 15.25% increase in total ballots cast, and Ralph Nader, who was on the ballot in 2000, was not on the ballot in 2004. Bush got all the new voters, all the Nader voters, and some of the Gore voters, or the equivalent, or else the official results are not true and correct. We audited nine such precincts, where Kerry got fewer votes than Gore. In these precincts, it was the voter signature books that told the tale. In eight of nine precincts, there were too many absentee ballots and too few regular ballots compared to the numbers of voter signatures in the book. In the ninth precinct, Van Wert 2-C, there were too many absentee ballots and too many regular ballots. This is clear evidence of after-the-fact ballot substitution. The ballots, as punched, did match the official vote count, or nearly so, but the breakdown of regular and absentee ballots was never correct. We also found the names of 75 “provisional voters,” all in the same handwriting, in the books for nine different precincts in six different wards, with none of these voters having signed in at the polls. This could not have been done by a poll worker, as no one can be in so many places at once. This is clear evidence of ballot box stuffing, consistent with an after-the-fact attempt to match an inflated vote count coming out of Chattanooga.

In Miami County, which used optical scanners, the number of ballots cast in the official results failed to match the number of actual voters in the poll books in every single precinct. Never was the official number of ballots cast correct. In Piqua 3-A, there were 32 provisional ballots missing. In 53 of 82 precincts there were too many provisional ballots, 238 altogether. In 74 of 82 precincts there were absentee ballots missing, 447 altogether. There were 567 “recount remakes,” substitute ballots filled out by the Board of Elections in preparation for the rigged “recount” staged in December 2004; 79 of them, or 13.9%, showed no mark for president, compared with a countywide rate of 0.67% for undervotes and overvotes combined, the lowest of any county in Ohio. (Statewide, the official percentage of uncounted ballots was 1.65% (94,535 of 5,722,443). The ballots for the sample precincts selected for the hand count were sorted in advance according to presidential candidate, and still the hand count did not match the official results. All of this is indicative of after-the-fact ballot alteration to match an altered vote count coming out of Chattanooga. This would have been easy to do. The election records were not stored in a secure, double-locked vault, as required by Ohio law. They were stored in the Power House across the street from the County Court House, and numerous persons had the key.

In Cuyahoga County, a variety of methods was used to suppress and alter the vote count. One was made possible by “ballot rotation.” In Ohio, according to law, the sequence in which the candidates’ names are listed on the ballot must rotate from precinct to precinct. In urban counties there were many “multiple precinct polling places.” In Cuyahoga County, in at least 41 precincts at 17 polling places, all of them heavily Democratic, voters were allowed to take their punch card ballots to the wrong voting booth, intended for a different precinct, in which case their votes were shifted to candidates not of their choosing – most notoriously in Cleveland 4F and 4N, where 379 Kerry votes and 20 Bush votes were shifted to Peroutka, Badnarik, and Nader. This had nothing to do with SMARTech. These ballots were counted exactly as punched by the voters, erroneously, through no fault of their own. These “mistakes” were by design, and we know this because a precinct judge was overruled by a mysterious “inspector” who instructed her to allow voters to “vote at any vacant booth.” At these “multiple precinct polling places,” if voters stood in the wrong line to begin with, their names would not appear in the poll books, and they were given provisional ballots, which then were not counted because they had voted in the wrong precinct, albeit at the right polling place. In Cuyahoga County, 7,450 of 24,788 provisional ballots (30.05%) were ruled invalid and were not counted – the second highest ratio in the state of Ohio. None of this had anything to do with SMARTech.

Cuyahoga County also had the highest number of voters purged from the voter rolls – 168,169 names, or 19.44% of the electorate. In Cleveland alone, 63,721 names were purged, or 24.93% of the electorate. Most were names of voters who had died or moved away – but not all. Officially, Kerry won Cleveland with 83.36% of the vote. For every six persons unable to vote for having been purged from the rolls, four votes were shaved from Kerry’s margin of victory. Troy Seman has shown that the ratio of voters purged was strongly correlated with the percentage of the vote won by Al Gore in the 2000 election. This strongly suggests that the most heavily Democratic wards in Cleveland were targeted for selective purging, which would be a violation of the Voting Rights Act. Again, this was a local operation, having nothing to do with SMARTech.

In fourteen Cuyahoga County precincts, we photographed all the absentee ballots. In eleven of these precincts, more absentee ballots existed and were counted than the total number issued to and returned by the voters. This is proof of absentee ballot box stuffing, most likely on behalf of George W. Bush who, in these fourteen precincts, received only 6% of the vote at the polls, compared with 20% of the absentee ballots. Again, this was a local operation, having nothing to do with SMARTech.

As stated above, nearly two-thirds of the uncounted punch card ballots in eight urban counties were due to multiple punches for president, resulting in an “overvote.” In Cuyahoga County, in 46 precincts audited, the ratio was 45.5% (353 of 775). Unlike the other seven counties, however, we did have the time to photograph every ballot issued at the polls. The ballots in the audited precincts of Cuyahoga County almost always matched the tabulator count. There is absolutely no evidence of any after-the-fact attempt to get the ballots to match a fraudulent vote count. Rather, the vote riggers utilized numerous methods to alter the vote count before the ballots were run through the tabulators.

In summary, we return to the question of whether the evidence of fraud uncovered by our investigation was related to the outsourcing of vote counting to SMARTech at 11:14 PM EST on Election Night. The vote rigging in Warren and Cuyahoga counties were strictly local operations, taking place before the punch card ballots were run through the tabulators. In Clark, Hamilton, and Montgomery counties, we never had the chance to look at ballots actually punched for Bush or Kerry, so we do not know if those ballots matched the official count. In Lorain County, we were not allowed to look at any ballots at all. In Mahoning County, with its touch screen voting machines, there were no paper ballots to look at. But in at least six counties, the methods of fraud we observed may well have been related to SMARTech. In Geauga and Delaware counties, the number of registered voters was artificially inflated, thus opening a window of opportunity to artificially inflate the vote count; we never asked to look at the ballots in Geauga County, but in Delaware County we found 359 consecutive ballots punched for George W. Bush. In Clermont County, some of the methods of election rigging were local operations, most notoriously the “stickered ballots.” But, in Clermont County, as in Butler, Miami, and Van Wert counties, we found evidence of ballot box stuffing, ballot destruction, ballot alteration, and/or ballot substitution in almost every audited precinct. We believed at the time that we were observing an after-the-fact attempt to get the ballots to match a fraudulent vote count, and we are more convinced of it now.


Another interesting timeline has to do with the CNN exit polls. We have two exit poll “screen shots.” Many researchers used only the first one because they liked its numbers. This is unfortunate, because it is the comparison between the two that tells the tale. #1 12:21 AM EST 1963 respondents Kerry 52% Bush 48% #2 1:41 AM EST 2020 respondents Kerry 48.5% Bush 51% Democratic Underground

In exit poll #1, Kerry had between 51.5% and 52.5% of 1963 respondents, or 1011 to 1031. Bush had between 47.5% and 48.5% of 1963 respondents, or 932 to 952. In exit poll #2, Kerry had between 48% and 49% of 2020 respondents, or 970 to 990. Bush had between 50.5% and 51.5% of 2020 respondents, or 1020 to 1040. It is not mathematically possible for 57 additional respondents to account for this shift. Not from the raw data. No matter how you look at it, Kerry had fewer supporters among the 2020 respondents at 1:14 AM EST than he had among the 1963 respondents at 12:21 AM EST, and Bush gained more than 57. This is proof that the polls were “adjusted” at 1:41 AM according to criteria different from those utilized at 12:21 AM. One might question where the exit pollsters found 57 additional respondents in the wee hours of the morning. One might also question whether the exit poll was altered to match an altered vote count. A precious few election observers did notice, right away, on Election Night, that 57 additional respondents were not enough to account for the change, and for this they should be commended. This is what we should be doing – auditing our elections.

The exit pollsters themselves discredited their own work and locked it away as “proprietary” data, contrary to standard scientific practice to subject one’s work to peer review. Warren Mitofsky, known as the inventor of the exit poll, refused to release precinct-level exit poll data from Ohio to researchers who wished to compare them to the official election results, which they maintained were fraudulent. Mitofsky's public explanation was the “reluctant Republican responder” hypothesis. From my own experience I know that this is often true. Republicans may be less willing than Democrats to cooperate with exit polls, in which case the raw data must be adjusted accordingly. But such adjustments cannot be made on Election Night – not without knowing the true ratio of Republicans and Democrats who actually voted at the polls. And no “adjustments” of exit poll data should be considered legitimate unless subjected to public scrutiny. One cannot claim that their work is “scientific” without adhering to scientific method. Any attempt to do so, especially among those educated in scientific methodology, is deliberate fraud. Just as on a math exam, you must “show your work.”


Richard Hayes Phillips was the leading forensic investigator of the 2004 presidential election in Ohio. He and his assistants photographed 126,000 ballots, 127 poll books, 141 voter signature books, and other election records, from 18 counties in Ohio. His analysis, and the raw data upon which the analysis is based, are presented in his book, "Witness to a Crime: A Citizens' Audit of an American Election." The book comes with a CD containing 1200 of the most telling photographs. His book is available at

Witness to a Crime Troy Seman has fifteen years experience working in Information Technology and dealing with computer security issues. He was Phillips’ most trusted adviser during the investigation of the 2004 election in Ohio.