The sample is D+5 though? And the sample for NBC was D+3. So that doesn't really jive.
The gist of the argument is that you can't just look at D-R self-ID margin alone, but also have to consider "independents". Given the basis of partisan non-response not inherently just applying to those who identify as partisan (because the vast majority of self-identified "independents" are indistinguishable from base voters in one of the two major parties, except public voter files are often useless in identifying them as strict-D or strict-R if you're a national pollster), the argument is basically that Trump independents and/or leaners are far less likely to answer surveys given the walloping he's had recently relative to Biden independents/leaners.
If such applies, then even the countermeasures that pollsters use (i.e. continuing to sample "under-represented" groups) won't yield an effectively different result based on how they categorize voters. That's because they'll keep sampling "independent"/leaning/swing voters and finding the same result with a partisan non-response bias.
One can look at this poll and see that Trump's support share really hasn't collapsed all that much: Trump dropped 2 points and Biden gained 6 points relative to the last CNN poll. This isn't inherently how swing voters behave, even this close to the election; they tend to go from D-to-undecided or R-to-undecided rather than making an absolute leap. If anything, CW would suggest the reverse for a 16-point margin: that Trump should've collapsed by 6 points and Biden would have gained 2 (i.e. 53-37).
There'd be an argument for this not being an outlier and/or not being subject to partisan non-response biases had Biden been posting low-double digit margins all month long and gradually growing, but alas.
