3) Now let's consider some flaws in their analysis.

A general problem is the mixing and matching of papers in ways that are arguably incompatible - different categories for measurement, different methodologies, etc.

Look at the details for HCQ, figure 5. This uses a scale from 0 (HCQ is beneficial) through 1 (no effect) to 2 (HCQ is harmful). It colours beneficial in green and harmful in red. But lots of this is wrong. Take a study that scores 0.97 on their scale. This is not statistically significant - it is so close to 1 that it cannot meaningfully be called beneficial. The same goes with 1.05 harmful - it's indistinguishable from no effect. Check many of these out, and these papers will not say they found HCQ was effective, their p values show so, and they clearly say so. They should therefore not be counted that way.

They've included confidence intervals (CI), in the horizontal lines around the dots. The confidence interval reflects the fact that studies are necessarily approximations of truth, and how likely they are to be accurate. So a 95% CI means that they believe there is a 95% chance the truth lies in that range based on their results. CI is not the same as a p value, but the concept is similar. A better way to interpret the data in that figure is that anything with a CI that crosses the value of 1 they cannot be confident is beneficial or harmful, because "no effect" is within the confidence value. Suddenly, that data starts looking very, very different and it becomes clear the majority of studies do not clearly find HCQ is beneficial.

Now let's go to figure 4, and let's take pre-exposure prophylaxis. They In one graph, they attempt to argue there is a 1:1000 likelihood HCQ is not beneficial, and there's a trendline indicating this. It is instantly observable that this calculation is fake. Essentially, every "positive" finding is used to make the claim HCQ is effective more reliable, and every negative one the opposite. However, here two things are clear. Firstly, studies that are counted as "beneficial" are not. Gendelman, Konig, Macias, Gianfrancesco, Gendebien, etc. You can see they do not show benefit, because of p values, although the minimal effectiveness should also give it away. Yet they are used to argue HCQ was beneficial for this trendline. Secondly, we see weighting has not been applied. Rentsch, for instance, was a massive, major study with tens of thousands of participants, about a sixth of all participants if we were to add all the total from all studies. It should have a massive effect on their trendline. (They've screwed the numbers up on the n values in figure 5 - sloppy.) Huh, the first, has 65,000. These two studies alone are half of all the participants in the total PrEP group, and neither show a benefit for HCQ. But individually they seem to carry the same weight as a study of about 100 participants. Nor does the robustness of the study design seem to factor in. Thus this argument for likelihood is completely false.

This is simply a taste of how inaccurate their analysis is. It is incredible unlikely that an honest, competent scientist who works in the field would make these sorts of errors, still less so a team (as they claim to be) who would have the advantage of cumulative expertise.

Given their claims of expertise, I am unfortunately forced to conclude that the creators of this site are intentionally dishonest.