Rejoinder:
Effective treatment is considered a public health intervention if it prevents the spread. Otherwise its medical treatment. This is why vaccination is far superior to post-infection treatment.
Assessment:
I'm deeply suspicious of this website. It uses pooled effects across several studies, which is an improper approach for this type of meta-analysis. For example, it lists hydroxychloroquine as effective. Some initial underpowered studies showed potential, but further research showed it was not effective.[0]
Items:
[1] By forcing a random effects design, you may give inappropriate weighting to under-powered studies showing an effect (low power results in higher effect sizes). 10 studies of 10 people each, with effect averaged via pooled regression, may well show a pooled effect significantly larger than one study with 100 people.
[2] People not deeply familiar with the treatment literature (such as you, me) cannot confirm that the meta-analysis is appropriately comprehensive (they claim to scrape paper sources, but not the inclusion criteria). Since this is a random website presented without context, without bonafides, and so forth, even if the information presented is real it may be cherry picked. Here is the documentation of some of their exclusion protocol -- note if they were intending to be persuasive instead of objective, exclusion of negative meta-analyses would be intentionally phrased objectively: https://hcqmeta.com/#exc
[3] Differing definitions can impact results. "Early treatment" is prophylaxis typically, but why the individual ends ups in a treatment or control group (and what the term means) may be systemically biased depending on the treatment standard of care adopted locally. The differences may not be random, resulting in selection bias (and other forms of bias, but simply put not an apples-to-apples comparison). And because we're dealing with people with will, the uncertainty compounds. By way of example, prophylaxis "treatment" group might be fully self-selected people who are just scared they may have been exposed, and a "control" group (from a matched pair example selected on few if any demographics) that was known to have the disease -- causing immediate positive bias. RCT designs can help with this approach, but again showing solely the pooled effects instead of reporting fixed effects assessments and other regression output is in err.
Conclusion:
So, for these reasons and because the source actually matters in an era of mis-/dis-information, I remain highly suspicious.
Notes
[0] Economics is facing its own replication issues, and the paper is an important read for why statistical power matters: https://onlinelibrary.wiley.com/doi/full/10.1111/ecoj.12461