that's the billion dollar question. literally, if you can solve that problem, then it's worth billions, as you'd be able to replace the indeeds of the world. i'd strongly contend that no one has even come close to actually solving this problem, which is probably in firmly the realm of P=NP.
my (now failed) startup in this space attempted to answer it for less variable jobs (hourly work), where we could tease out more signal from the noise, but even that had lots of unaccountable variability.