Don’t like polls from media outlets and academic institutions? Meet Remington Research, which regularly conducts polls for Axiom Strategies, a Republican-oriented consultancy. Both were founded by Jeff Roe, a longtime GOP strategist who has worked at times for Mike Huckabee and Ted Cruz, specifically on his presidential bid this cycle. Axiom has a very good success rate in Congressional elections (81% of its clients have won their elections as of early 2016), and Roe got a great deal of the credit for Cruz’ surprise win in the Iowa caucuses in February.

Remington has turned its focus to swing-state polling, and finds Donald Trump up in one key state and within striking distance of Hillary Clinton in another. Both, however, raise questions about reliability.

Let’s start with Ohio, where Remington finds Trump leading by four points, 46/42. In order to get there, Remington starts off with a poll sample that has an R+9 partisan composition (34/43/23). That’s, um, a very optimistic election turnout model for Ohio. George W. Bush won the state in 2004 by three points while the split was even at 37/37/26. In 2012, the most recent presidential election, Ohio had a D+7 turnout model at 38/31/31. Even with John Kasich running against the utterly inept Ed Fitzgerald in the 2014 midterms, the GOP edge in turnout was only four points at 32/36/32.

The issue of partisan composition of samples gets overused as a credibility argument, but this is notable for three reasons. One, it posits a sixteen-point swing in the gap in turnout model from the previous presidential election, which seems highly unlikely. Second, this is precisely the issue that Trump’s supporters make when dismissing other polls. Third, with an R+9 turnout model in a state that traditionally leans GOP, we should see a much larger lead if Trump really is generating that kind of turnout among Republican voters — especially if he’s winning Hamilton County (Cincinnati) by 13 points and getting 19% of the African-American vote. Those should add up to a Kasich-scale landslide, not a just-outside-the-MoE lead.

Pennsylvania’s results seem a little more credible, but still have issues. Hillary leads 45/42, but only leads in Philadelphia by 56/32, a county won by Obama 85/14 on his way to a 5-point win in 2012. Again, Remington has Trump winning a much higher percentage of the African-American vote than any Republican has in decades (29%, compared to Romney’s 7% and George W. Bush’s 16%). Those two data points alone suggest that Trump should be doing better than 42% overall, a lower figure than either Romney or Bush eventually got. The partisan breakdown in this poll is much more reasonable (D+6, 48/42/10), but Hillary’s thin six-point margin among women seems a little suspect, too.

Could these polls reflect a reality that has escaped other pollsters? Sure, and Remington does get included in aggregation at RCP and FiveThirtyEight, so there is general acceptance of their results. Axiom doesn’t succeed by giving its clients bad data and bad strategies, either. However, it’s still a Republican shop, and publicly released polling might have other purposes than just dispassionate analysis, just as it would for Democrat-aligned pollsters. The data itself suggests that readers should have several grains of salt on standby.