The 2004 Solution: a Manual Parallel Accounting Control System

Archive record — page as published on ballotintegrity.org, October 2004.

The page the 2004 site called its "2004 Solution" set out a proposal known as a manual parallel accounting control system, or MPACS. It is the page for which this domain was most often cited, and the idea deserves a careful, neutral explanation — partly because it was widely misdescribed at the time, and partly because the question it addressed has since been answered far more rigorously than the proposal itself managed.

The problem the proposal addressed

The starting observation, as published in 2004, was that public discussion of electronic voting had concentrated almost entirely on the touch-screen machine — the direct-recording electronic device a voter uses in the booth. The page argued this framing was too narrow. Even in jurisdictions that used paper — optical-scan ballots marked by hand, or punch cards — the totals from individual precincts were commonly accumulated on central tabulation systems supplied by the same small group of vendors. If the concern was that a tabulation process could not be independently inspected, then the concern applied to the aggregation stage regardless of what the voter had touched.

The second observation was about recounts. The page argued that an occasional recount, triggered by a close margin or a candidate's request, is not an audit. A recount examines one contest, after the fact, in circumstances that are usually adversarial and always exceptional. It provides no routine assurance about the overwhelming majority of contests, which are never recounted at all. Something that ran alongside the ordinary count, every time, would be a different instrument.

The third observation was about timing. The page was published in the autumn of 2004, a matter of weeks before a general election. Whatever the merits of longer-term technical fixes — certification standards, software inspection regimes, equipment replacement — none of them could be delivered in the time available. The project therefore proposed something that required no new equipment, no procurement and no software: people, paper and arithmetic.

What MPACS actually proposed

Stripped to essentials, the proposal was to run a manual count in parallel with the official machine count, using the same ballots or the same voter-verified records, and to treat the comparison between the two as the control. The elements were:

• Parallel, not replacement. The machine count would proceed exactly as normal. The manual process ran alongside it as an accounting control, in the sense a bookkeeper would recognise: an independent second record produced by a different method, against which the first can be reconciled.
• Local and immediate. Counting would take place at the precinct, at the close of polls, before ballots or memory devices were transported anywhere — so that the reconciliation happened while everything was still in one place and under observation.
• Publicly observable. The count would be open to observers, and precinct totals would be posted at the polling place, creating a public record independent of any transmission or aggregation step.
• Narrow in scope. The project accepted that hand counting an entire ballot with dozens of contests and referendum questions was not feasible at scale on short notice. The proposal was confined to a small number of federal races, which is what made the arithmetic manageable.

The operational version of that idea, worked out in detail, is on the plan page; the practical questions volunteers were asked about it are answered on the FAQ page; and the equivalent recommendations for counties using optical-scan equipment are in the accountability procedures.

The public opinion argument

The page also made an argument about confidence rather than accuracy, and supported it with contemporary polling. It cited a survey published in September 2004 reporting that roughly four in ten Americans were worried about electronic voting machines, with about 42 per cent expressing concern about tampering and about 38 per cent about accuracy. Those figures are recorded here as a 2004 third-party statistic, quoted as the page quoted it. They are not current public opinion, they are not verified here, and they should not be read as describing attitudes at any later date.

The use the page made of them, though, is worth noting, because it is a distinct argument: that a counting process which cannot be observed will be doubted whether or not it is accurate, and that observability is therefore worth something on its own terms. That claim has aged better than most of the technical detail around it.

In context: what auditing looks like now

The gap between MPACS and modern practice is instructive, and it is a gap of statistical sophistication rather than of principle. A parallel manual count of every precinct is enormously expensive in volunteer labour and produces no measure of how much confidence it justifies. The instrument that emerged instead is the risk-limiting audit: a procedure that draws a sample of paper records, sized according to the reported margin, and continues sampling — up to and including a full hand count — until the evidence supports the reported outcome at a stated statistical confidence level, or the outcome is corrected.

That approach delivers what the 2004 proposal was reaching for, at a fraction of the effort, and it does so with a defined and reportable error bound. It also depends on exactly the precondition the project spent 2004 arguing for: a durable, voter-verified paper record for every vote, without which there is nothing to sample. The National Conference of State Legislatures maintains a current summary of which states require post-election audits and in what form, and the U.S. Election Assistance Commission publishes related research through its studies and reports programme. The technical standards underpinning the systems being audited come out of the NIST voting technology programme.

For a contemporaneous assessment from the same period, the Brennan Center for Justice published a technical study in 2006, The Machinery of Democracy: Protecting Elections in an Electronic World, which examined threats to the voting systems then in use and concluded that routine auditing of voter-verified paper records was among the most effective available countermeasures. Those are that report's 2006 findings, attributed to its authors; they are cited here because they are the nearest rigorous treatment of the question this page raised two years earlier, not as an endorsement of anything on this site.