Would anyone know exactly how TC is calculated? I have a general grasp of it as some gradient of the meta model return performance with respect to my stake size, if that is correct.
For example, I remember there was some illustration of how the meta model is combined from our predictions with our stakes. Could we view the detailed graph so we can analyze the back propagation? I don’t know what I’m looking for, or if I will uncover anything. I just want to understand TC better so I can think how to best optimize for it. The exact forward pass and backward derivative in math formulas would be ideal.
There have been discussions already to target FNCv3 or have low correlation with meta model. I think those are correct given the empirical evidence. I imagine they somehow target more obscure areas to provide the meta model with better balance. But I do not think it would be generally true all the time.