Projector point isolation : Cosine distance change and explained variance decrease

Hello,
This issue is a request for an explaination about two points in Tensorboard's projector :

1 - Cosine distance changes when isolating a selection of points

Cosine distance seems to change when zooming on points. The right side column showing the nearest neighours has the mention 'in the original space'. I think I'm not the only one who also interpreted it as 'the cosine distance is also calculated in the original space', so I think it should be mentioned that it is not the case. The cosine distance seems to have been mapped from it's original interval to an interval somewhat close to [0, 2], but I couldn't find the formula for this change anywhere.
To sum it up:

• The change occuring in the formula should be mentioned and documented
• We should have a button to enable/disable this modified cosine distance

2 - PCA displays an explained variance of 100.0%, but it decreases when isolating points

Initial PCA with 100% explained variance shows a superposition of points with a cosine distance value of 0.000. When selecting and isolating these points, the superposition is no more, and the explained variance goes down to 92.6%. The superposed points are 20 in a 450 points representation, and there are many of such superposition cases in the same embedding (hence the fact that 100% variance explained is a bit shady.
Here is a link to my Tensorboard files folder.
An example of the case of superposition can be found in the embedding Embedddings/t_embedding if you look for the point labeled technicien assurance qualité.

Thank you for your time, and pardon my mistakes if I submitted this first issue in the wrong format.

Tensorboard version : 1.8.0
Tensorflow version: 1.6.0
Python Version : 3.6
OS : Windows 10

Start in TensorBoard's Projector with the supplied embedding and reproduce the cosine-distance and PCA explained-variance changes when points are isolated. Compare the initial view with the isolated selection, then document the distance mapping and explain the variance change; clarify whether a control for the modified distance is still required.