When you optimize for, say, the number of publications, you encourage aspiring professors to vomit out more minimal publishable units and self-promote more loudly rather than thinking interesting thoughts and writing them down after some rumination.
The property you actually care about, the interestingness and quality of the research, is a latent property. You cannot optimize for it by optimizing any of its indicators; they are only noisily related, and intervening on the indicators will only break the connection between the indicator and the latent interestingness.
When you optimize for, say, the number of publications, you encourage aspiring professors to vomit out more minimal publishable units and self-promote more loudly rather than thinking interesting thoughts and writing them down after some rumination.
The property you actually care about, the interestingness and quality of the research, is a latent property. You cannot optimize for it by optimizing any of its indicators; they are only noisily related, and intervening on the indicators will only break the connection between the indicator and the latent interestingness.