I was playing around with a few popular alternative data APIs.
Some of them are:
I backtested different variations of them on 3-5 years of historical data.
I also calculated benchmarks for them.
I sorted them based on their performance in relation to the benchmark.
The results are below.
It seems that most of those strategies are there simply by chance, if you test a sufficiently large number of strategies some of them will just work.
Some of them definitely don't work, US Congress Members have 45 days to make their trades public.
Some of them seem to work, for example, GitHub commits. I do understand that the quality of the code cannot be judged by the number of commits, but perhaps developers are the first ones to know that the project is not going well and they lose interest.