Manufacturers can use an outdated version of the EPA test and then apply a fudge factor of their choosing.
Technically not quite correct. Actually they can only use the higher value of the fudge factor if they do the more comprehensive set of tests. If they only do two-cycle testing (UDDS and HWFE) then they must use 0.7 to scale the weighted (55% and 45%) results of these tests. If they do test all five cycles, then they can (if they wish) scale their UDDS and HWFE results with a scalar factor (which so far has gone as high as ~0.76) which is based on a set EPA formula, using the results of all five types of cycles (adding US06 (“high” speed), 20F FTP (cold), and SC03 (air-conditioning)). At least that is my basic understanding.
So if you do better on high-speed/cold/AC, you can get a higher scalar for your weighted UDDS and HWFE cycle results. And thus a better range result.
A few manufacturers just choose to use 0.7. For example, VW, as I recall, lol. Also some other American manufacturers.
Of course, no matter what is done, the result is not very useful to the end customer, with current EPA regulations. It does not predict realistic highway range, which I would argue is most relevant.
I bet @AlanSubie4Life could write a 127 page document of criticism.
Would just need to cut and paste from this website.
Anyway, perhaps if the EPA used machine learning they could get to the bottom of their range fiasco. It should be punished, just like Twitter.