A single metric can never capture the real world nuance.
Data scientists and finance people typically look at a portfolio of metrics that are all correlated in different partial ways with what they actually care about, and then triangulate.
In that regime, there must be someone to ultimately make a judgment call about which metrics matter at a given time.
That person making the call can draw on their nuanced experience and context, and should have long-term alignment and responsibility for the outcome.
The tech industry seems relatively naive in this domain.
The tech industry tends toward "figure out a single objective metric and simply make it go up."
There have been formative experiences like when the newsfeed switched to algorithmic, where users protested loudly, but the metrics showed that users loved it.
This led to a mindset in the industry of "the users don't know what they want so just laser focus on the metric."
But that's the wrong lesson!
The metrics can only show what the limbic system wants, not what the user wants to want.
The search for a single "objective" metric precludes capturing complexity and nuance.
Each individual metric is a half-truth.
Which metric to pay attention to in a given moment is a judgment call… you can either be explicit about that or ignore that.