The vote took place on Oct. 10, and P&G has asserted that the “preliminary results” show that it prevailed. But the P&G tally isn’t based on an actual count of all the votes cast. The consumer products titan isn’t following the standard practice of just releasing the tally of the votes it’s sure of. Instead, it relies on half-a-dozen estimates and assumptions for the substantial number of votes it hasn’t yet seen to arrive at an estimated final count.

“P&G’s reporting of the vote is unusual,” says Bruce Goldfarb, founder and CEO of proxy solicitation firm Okapi Partners, which was not involved in the battle. “But there have been very few proxy battles in the last decade that have involved a shareholder base the size of P&G’s.”

P&G’s official, estimates-based reporting has been mostly ignored in the press. But it’s highly significant, because even its preliminary numbers show an extraordinarily close contest. According to the company’s tally, Peltz lost by 6.15 million votes, or just 0.2% of P&G’s 2.65 billion eligible shares, and 0.31% of the votes actually cast. Peltz received 48.6% of the vote to his opponent’s 48.9%. Peltz’s Trian Fund Management, owner of $3.5 billion worth of P&G shares, is calling the vote too close to call, and has assailed P&G for announcing totals based on “estimates and incomplete information.”

A guesstimate on votes from the little folks

While it’s unlikely that the final, certified tally will put Peltz on the board, Trian still makes a good point. On Monday, six days after announcing its win at the annual meeting, P&G submitted an 8K regulatory filing that for the first time, announced the preliminary numbers. But P&G relied on estimates for a crucial voting block: Main Street individual-investor types who own stock through brokers and custody accounts administered by banks. And P&G, a favorite of small investors for decades, has far more of those individual shareholders than most big companies.

The company relied on estimates because votes from those individual, “registered” owners are submitted only to one side. The pro-P&G votes on blue proxy cards are seen only by its camp, and the votes for Peltz on the white cards are sent only to the Trian side. Naturally, each side knows the number of registered shares outstanding, so that it can estimate from its own count how many votes went to their opponent. Still, lots of uncertainty remains. Neither side can be sure of what portion of the registered holders cast ballots, including the last-minute paper proxies submitted at the annual meeting. Also clouding the outcome: Unlike a political election, proxy contests allow shareholders to vote multiple times. It’s only the last of their votes that counts. Hence, a pro-P&G ballot that the company has in hand could be trumped by a later proxy from the same shareholder who’s switched to Trian.

In its 8K, P&G acknowledges that the results are “preliminary estimates only.” Indeed, what’s unusual is the frequent reference to “estimates.” The numbers that P&G has actually seen are those collected by its two proxy solicitors, including the votes sent electronically by big institutional investors. For the ballots from custody and brokerage accounts, it relies on estimates––and takes the unusual step of translating those estimates into hard figures and projected results. In a footnote at the bottom on page 2 of the 8K, P&G offers no fewer than six areas in which it has deployed “estimates.” The list includes votes “on behalf of custodian banks and brokerage clients,” “proxy cards voted at the annual meeting,” and “revocations” of votes on each side based on which combatant received the last vote from investors who switched sides.

It’s highly instructive that in Trian’s losing proxy contest versus DuPont in 2015, the chemical giant took a more conservative, data-based approach to announcing the preliminary results. In its 8K dated May 19, 2015, DuPont never employed the term “estimates.” It unveiled only the tally “based on information available to,” meaning actually counted by, its proxy solicitors, adding that the information is “incomplete” and citing all the categories of votes its solicitors hadn’t seen—without making any estimates of how those votes were divided.

Where does the battle go from here? Both sides will submit the ballots the other camp hasn’t seen to IVS Associates, the independent inspector of elections that will tabulate all of the votes and certify the winner. “I don’t ever recall seeing a footnote like the one in P&G’s 8K,” concludes Gardner Davis, a partner at Foley & Lardner, a law firm that has represented both activists and targets in proxy battles. “This is so much of a high-stakes battle. I think they were under tremendous pressure to declare victory.”

P&G would have been better served reporting what it knows, not what it projects. Shareholders are waiting to see if its estimates reflect the actual voting, just as they’re waiting to see if its forecasts of a major turnaround turn into the buoyant comeback the company promised in its campaign to defeat Peltz.