Jazz blogged it this morning but it’s newsy enough to warrant a follow-up, as it represents the first strong sign that Hillary’s convention bounce — and/or the backlash from the Trump/Khan dust-up — is fading. Jazz makes two good points for why we should believe the result. One: Reuters did detect a big lead for Hillary a few days ago. If it hadn’t, you could challenge the poll as an extreme outlier given that every other survey this past week captured a spike in her numbers. Reuters captured it too, though. And now they see it going away. If you’re looking for evidence that she got a dead-cat bounce and the race is soon to tighten again, this is it. Two: This is a poll of likely voters, not registereds or adults. A tighter voter screen should produce a poll that more accurately reflects what the vote will actually look like on Election Day. Every other poll this week, with Clinton piling up big leads in survey after survey, has been a poll of registered voters, not likely voters. Except one: The LA Times daily tracker is also a poll of likely voters, and that one also has the race much closer than everyone else (Clinton leads by just four-tenths of a point). There is reason to believe that this data is closer to the true state of the race than the ones showing Hillary up double digits.

But there are a few quirks that should be noted. Open up Reuters’s graph of its results, which can be customized to show trends among different groups via filters at the bottom, and follow along.

1. Traditionally it’s been common for likely voters to vote more Republican than registered voters do. That’s what you see in this poll too: Hillary leads by nearly six points among registereds but by less than three among likelies. This year, however, there’s a caveat. Some polls have showed Clinton doing better among likely voters than she does with registereds. Nate Silver tried to puzzle that out a few months ago and guessed that it’s an artifact of Hillary performing much better with college-educated voters than Democrats tend to do, and college-educated voters tend to be relatively high-turnout voters. You’re not seeing that effect here. That doesn’t mean Reuters’s numbers are wrong but it’s worth noting that they’re not detecting the same phenomenon as some other surveys are finding.

2. Normally when you want to understand movement in Hillary’s polls, you look at women voters. This week her lead in other surveys ballooned from 13-14 points among women pre-convention to 20+ points now. Reuters has always tracked her a bit lower among women but they saw her lead balloon for a few days too this week within that group before reverting to pre-convention levels yesterday.

It’s not women who account for most of Hillary’s dead-cat bounce in Reuters, though. It’s men. Check out this graph of male likely voters. The red line is men who support Trump, of course, and the blue is men who support Clinton.

That’s quite a trendline. On July 22nd, three days before the Democratic convention, Trump led men by 10; eight days later, Clinton suddenly led men by eight; five days after that, Trump suddenly leads by six points again and rising. Going from +10 to -8 to +6 again in the span of just two weeks — two terrible weeks for Trump, it should be added — seems weirdly volatile to me even in the context of a post-convention bounce for Hillary. An 18-point swing followed by a 14-point swing? In 14 days? But your mileage may vary.

3. On July 22nd, Trump led Clinton among whites, 53/30. That 23-point spread is plausible given that Romney defeated Obama by 20 among whites, 59/39. Trump has greater appeal to working-class whites than Mitt does, therefore it’s easy to believe that he might be building on Romney’s lead. (On the other hand, Trump has been losing college-educated whites to Clinton in other polls, which means, if anything, that he’s probably behind Mitt’s pace with whites in reality.) After the Democratic convention, Reuters saw Trump’s lead among white likely voters narrow dramatically but today they have it sitting at … exactly 10 points, less than half of what it was two weeks ago. In other words, although Trump is counting heavily on whites to propel him to victory and although his lead among that group is a measly half of what it was for Romney in 2012, somehow he’s nonetheless trailing Hillary by a smaller margin (2.4 points) than Romney ended up losing to Obama (3.9 points). How can that be possible?

The answer: Lots and lots and lots of whites who, for the moment, aren’t supporting either Trump or Hillary. More than 20 percent(!!) of Reuters’s sample of white likely voters chose “other,” refused to answer about their presidential preference, or said they won’t vote this year. That’s ironic because, until a few weeks ago, Reuters had a big problem in its polls with people choosing “neither” as an option instead of Trump or Hillary. So they eliminated “neither” as a choice in order to force people to pick one of the two nominees. And yet they’re still getting tons of none-of-the-above type answers, which makes it harder to guess what the true state of the race is. Which way do those white voters eventually break? Do they come home to the GOP reluctantly or is their animosity to Trump so great that they tilt to Hillary? It’s hard to compare Reuters to other polls because, for whatever reason, they have more trouble getting their samples to indicate a preference, even if it’s a weak preference, between the actual candidates. One way to read the result here compared to the other polls this week showing Hillary far ahead, then, is that she has a small lead if you’re only looking at voters who are fairly strongly committed to their candidate. Add in the ones who are weakly committed and who would rather not even be polled on their choice and maybe her lead balloons. But then, that’s why Team Clinton is spending millions on a sophisticated ground game — to make sure the “meh, Hillary” voters are at the polls on Election Day too.

But there’s one last possibility that might explain why her lead is unusually small here.

4. My pal Karl pointed out a bizarre result in the Reuters numbers. When you try to look at Hispanic likely voters, you get … nothing. “There Aren’t Enough Responses To Give Sufficiently Accurate Results” appears in lieu of a graph. You get a similar message when you try to look at black likely voters, but you can see the effect of blacks on the results if you look at white likely voters first and then add blacks to your result. (The numbers move from a 10-point overall lead for Trump to a lead for Hillary of one-tenth of a point.) When you try to add Hispanics also, though, the overall numbers barely move. (Hillary’s lead grows to eight-tenths of a point.) Either Reuters’s sample of Hispanic likely voters is freakishly tiny or it points to that group being very evenly divided between Trump and Clinton, which simply isn’t true. According to the 2012 exit polls, Hispanics were 10 percent of the electorate and went for Obama by 44 points, meaning that they were worth 4.4 points to him nationally. Most polls this year show Trump performing comparably to Romney among Hispanics or worse. Maybe the Hispanic data in Reuters’s poll is priced into the topline result but it’s simply not showing up in the graph for whatever glitchy reason. But if Reuters really did dramatically undersample that group then obviously Hillary’s true share of the vote overall is several points higher than the 41.5 percent she’s pulling here.

Anyway. Whatever the lead is, it’s not prohibitive (yet). In lieu of an exit question, read Amy Walter: “Why the 2016 Race Is Not Over.”