Comparison between velocity length and the velocity grid results

[AndiMunteanu]

Hello again!

I have a question related to interpreting the velocity length.
The velocity_embedding function allows illustrating the cell-cell transitions on the UMAP reduced space. If I understand correctly, the arrow's length can be correlated with a strong transition.
I was hoping to quantify the weakness/strongness of a transition, and I ended up looking at the velocity_length, hoping that it might contain the answer to my question.
However, the two plots (the embedding grid and the velocity length) are incomparable. In some cases, the arrow is very short, but the velocity length is very high, or vice versa.

I looked further in the paper and the documentation and, if I managed to understand this detail, the transition matrix is based on the similarity performed on both velocity and difference on expression level, whereas the velocity length is calculated based only on the velocity vector.

Supposing my understanding is correct, which one of the two information is more reliable in understanding the speed of transition between cell states?
Also, would it be possible to have a function that could calculate the velocity length by using the difference in expression as well?

On a slightly different topic, how should the output of the velocity_confidence_transition be interpreted, given that it has both positive and negative values?

Thank you and please let me know if what I said doesn't make sense at all.