Author here, this is not my job. I disagree with a lot of your points; for example a sample size of 19000 is great for testing more than 1-2 hypothesis; you can generally correct for p-hacking by multiplying the p-value you get by the number of hypothesis you test. I didn't check p-values because I'm working with likelihood ratios (which is what LR meant, I agree I should have explained this more but this was a memey graph and I usually go into more detail in other posts with data I take more seriously) which as far as I can tell are better than P-values. LR=1 e18 means that the likelihood of the given correlation appearing in the data (i think it was 0.06 or something?) is 100,000,000,000,000,000 more likely than a correlation of 0. If you presented this as a p-value, this would likely look like "p=<0.01" which is... way less information than the likelihood ratio.

And p-hacking isn't really a concern if you're dealing with likelihood ratios of that magnitude. I'd start to be concerned for likelihood ratios of around 100 or so. For contrast, most "statistical significance" in most published papers has a rough equivalent of likelihood ratio of 20.

this is just one point, i don't have time to go into all the rest right now