HERE'S THE CONVENTIONAL interpretation of the most recent media recount of the Bush-Gore election: Bush would have won even if the U.S. Supreme Court had not stopped the statewide recount of undervotes ordered by the Florida Supreme Court. But Gore would have won a statewide recount (that he did not request) of undervotes and overvotes. This seems to confirm that the U.S. Supreme Court was wrong to intervene, since the system would have produced a Bush victory anyway, and it further seems to confirm that Gore "really" won. This interpretation is wrong, both in its factual premises and in its conclusions.

First, the media recount does not show Bush would have won if Florida's manual recount of undervotes had continued. What it shows (as was apparent at the time) is that Bush would have won such a recount conducted under standards applied uniformly within each county by counters who were screened for their political bias.

But that was decidedly not the process underway in Florida on December 12. Then recounting was being conducted by unscreened temporary workers supervised by partisan election officials. Nor had each county picked one standard in advance, and stuck to it. Palm Beach and Broward began by using the only preexisting written standard, namely, that there had to be some perforation of the ballot. But then, after early results showed this did not pick up many votes for Gore, they switched to a dimple standard. Later still, these counties decided to switch to a policy of exercising discretion over which dimples they counted. By the end, as Gore's counsel memorably conceded, the standard being applied varied from table to table.

Does it matter? The media recount confirms that it does. The media consortium--the New York Times, the Washington Post, the Los Angeles Times, the Wall Street Journal, the Associated Press, CNN, and four Florida newspapers--contracted with the National Opinion Research Center to examine all the uncounted ballots in the state. Yet even when a single standard was specified, the counters hired by NORC frequently disagreed in their ballot interpretation.

Although some accounts stress that the counters agreed on 96 percent of punchcard ballots, that 4 percent error rate greatly exceeded the election margin of .001 percent. This is rather like trying to recheck a microscope's measurement of an electron's width using the human eye and a yardstick. Moreover, the 96 percent figure is artificially inflated by agreements on ballots where there was no marking to dispute. On ballots where at least one counter saw a potential vote for Bush or Gore, the counters disagreed a third of the time.

Political affiliation mattered. Though the NORC counters were supposed to be impartial, Republican counters were 4 percent more likely than Democratic counters to deny a mark was for Gore. Even more striking, Democrats were 25 percent more likely to deny a mark was for Bush. This bias may well be utterly unconscious, but it remains a problem for any manual recount process.

Indeed, if this is the sort of accuracy one gets from an unhurried professional effort when counters are screened for bias and bound to the same standard, imagine the sort of inaccuracy that would have been produced by a rushed partisan set of counters each free to choose whatever standard he wanted. The U.S. Supreme Court was amply justified in putting a stop to it.

Critics of the High Court have argued that Florida's manual recount--while inaccurate, arbitrary, and haphazard--was not unconstitutional. Ronald Dworkin, for example, argues that the equal protection clause is violated only when state law creates "distinctions that put some citizens, in advance, at a disadvantage against others." But what made this process alarming was precisely that it did not set forth any objective standards "in advance."

Such standardless discretion in the hands of partisan county officials is worrisome because it allows them to engage in sub rosa discrimination against the opposing party about how (and indeed whether) to conduct manual recounts. Since without standards such discrimination is hard to prove, the best way to vindicate the constitutional right of equal treatment is to prevent partisan officials from exercising such standardless discretion at all. For precisely this reason, well-established Supreme Court precedent makes such standardless discretion unlawful if used to hand out parade permits or locate newspaper boxes. Why should the protection be any less when discretion is being exercised over the far more fundamental question of which votes to count?