Inspired by the discovery of the extremely hot variable DDE 156 (helium dwarf with a period of 0.6977 d) I decided to check if there are similar objects hiding in VSX under the RRAB mark. I have made a search for RRAB variables in the period range from 0.65 to 0.75 days, saved them as .csv file, copied their coordinates to plain text file, and then queried GALEX catalog in Vizier using "List Of Targets" tab. I have found several objects that are definitely not RRAB variables. Some of them are not variable stars at all, being rather extragalactic objects with (gamma-ray) bells and (radio) whistles. Surely enough, they were published in refereed journals, having been discovered by the professional surveys. If you have ever tried to submit a new variable to VSX, you know the reviewing process by VSX admins is much more stricter than by the journal referees and editors. As a result, VSX still contains low quality data provided by third-party publications in the sources like Astrophysical Journal (ApJ), Monthly Notices (MNRAS) and Astronomical Journal (AJ). Here are some examples.
https://www.aavso.org/vsx/index.php?view=detail.top&oid=287743 [MVD2012] 1004 with P=0.717799 d (QUEST, MNRAS 2012) or P=0.7826694 (PanSTARRS, AJ 2017) which is actually a quasar with z=1.2, variable NVSS radio source.
https://www.aavso.org/vsx/index.php?view=detail.top&oid=142636 [VZA2004] 484 with P=0.665114 d (QUEST, 2004 AJ) which is actually a galaxy (!!!) PGC 1150218.
https://www.aavso.org/vsx/index.php?view=detail.top&oid=480483 SSS_J034838.1-274913 with P=0.703296542 d (CRTS, MNRAS 2017) which is active galactic nucleus (AGN) at z=0.9875, radio source PKS 0346-279 and Fermi gamma-ray source.
https://www.aavso.org/vsx/index.php?view=detail.top&oid=289355 CSS_J072041.9+343745 with P=0.6741402 d which is not RRAB, but probably the reflecting hot subdwarf with (FUV-NUV)=-0.50. The period may be right, or it may be a daily alias at 0.402223 d.
What can be done to cure those misclassifications in VSX?
1) All objects with (W1-W2)>1 and (W2-W3)>2 should be verified for the extragalactic origin, that is matched with the catalogs of radio sources and gamma-ray selected quasars.
2) All variables of "blue" types should be cross-matched against the catalogs of infrared sources (2MASS, WISE), and all "red" variables (Mirae, SR, L) - against those of ultraviolet sources (GALEX).
3) Now that Gaia parallaxes are available for hundreds million objects, the absolute magnitudes of many variable stars can be verified by Vizier TAP and/or X-match queries. That can distinguish between the red dwarfs and red giants, or between white dwarfs, hot subdwarfs and blue giants.
As a by-product, several unusual variables will stand out inevitably - for example, white dwarf plus red giant systems (symbiotic variables) or WD+RD binaries, be it detached or interacting. The final step for verifying the true classification could be checking the archival light curves in CRTS, ASAS-SN and other databases.
It really sounds like a nice project which already gave interesting results as you have proved.
What to do? Submit revisions. People usually tend to submit new discoveries but do not realize that revising wrong information is so important.
Revising existing data takes much longer because you need to double-check, verify and at times make assumptions that the newly published data are better than the existing one. If not we wouldn't be able to handle massive datasets.
I was taking a look at the Pan-STARRS1 RR Lyrae list that you mentioned in your post. I saw lots of false alarms. It will be a real problem to update VSX with this list.
I am also working with CRTS updates for existing variables and I found that half of the published results might be spurious. I do as many assumptions as I can based on the quality and nature of the data but for a number of objects I have to check light curves to decide if an object need to be updated or not.
Inevitably there will be some revisions that will overwrite good data with bad one. It is impossible to avoid such circumstance when working with thousands of variables at a time. But we try to keep that at a minimum.
Submitting revisions for the ones you found would be a great starting point.