Cut-off criteria for dropping from model execution to linear inventory datasets

[cmutel]

Traversing value chains and running models gives a lot of flexibility, but is also more slower and more resource intensive than building a matrix and solving a linear system. We expect to have most calculations be hybrid, and to use both approaches. We will start with model execution and graph traversal, and drop down to the generic background matrix when a cut-off criteria is reached.

What does that cut-off criteria look like? We will have the generic background matrix available for analysis before the calculation starts, and can also run models ahead of time and cache their results.

The cut-off criteria could be impact-category specific, though this could multiple the amount of information needed dramatically. This is what e.g. Temporalis does, though it doesn't cache the cut-off value but computes it dynamically.

[cmutel]

@schoi839 notes that some models can produce negative results (carbon capture, heavy metal uptake by plants). Need to be careful as convergence to the total score can be noisy.