Engine sets floating point errors as loss

[mabijkerk]

Context

Running this query for a blank nl2023 scenario gives:

V(energy_blastfurnace_transformation_cokes,input_of_loss)
7.62939453125e-06

This value is the same for the start year and for the end year. The thing is that there is actually no input of loss on the node itself:

Steps to reproduce

When downloading the present or future year energy flows (complete) it turns out that a lot of very small decimal input of loss shows up throughout the graph. See here a selection for the present flows in the nl2023 scenario:

There are plenty of nodes with input of loss in MJ very close to zero.

Proposed solution

It seems that loss is used as a generic 'last degree of freedom' for the graph, both on the input and the output-side. More generally speaking, we have been aware that floating point erros are a broader issue throughout our calculations, see the linked issues.

We need a comprehensive plan to change how we set up our calculations in the model to deal with these floating point errors. The challenge here is that the issue is relatively small, whereas the effort required to fix it can be large. We need to assess whether changing this is worth the effort.