Lets see if I can explain the logic.
You have system you want to model. Every group tries different approaches, there are over 100. The models have spread. Since they all use the same inputs, the spread is a result of modelling choices and inherent uncertainty in the complex process. To reduce the spread ( structural plus inherent uncertainty) you have some options: run each model many times ( wall time kills you) or find a model selection criteria.
The first obvious criteria is using temperature as a constraint. Pick those models that get temperature correct. Then we you look at the ECS value of the selected models you get an answer constrained by matching temperature.
But temperature change is a part of ECS calculation and some models are indirectlyy tuned by temperature or closely related variables.
So you look for an EMERGENT property: That is a property or metric that is not directly tied to inputs or tied to temperature. You select models based on this emergent ( develops as a result of programmed physics) property, so in the end you are using one emergent metric to constrain a different emergent metric (ECS)
Interesting approach to reducing structural uncertainty.
The good thing is it focuses your attention on areas not directly related to your inputs ( forcings– the denominator of ECS) or temperature ( the numerator of ECS)