Here is a Fork with some examples of what I mean. Feel free to edit/reject if you want

Really nice kata. I suggest adding a few more sample tests (based on things I overlooked initially in my solution):

• One identical to the first test, but with the first shape coordinates in reverse order
• One where the two shapes create a "window" of empty space. (eg. a V shape, and a rectangle along the top)
• One where a line segment of the combined shape begins and ends at intersection points, not at vertexes of the input shapes

Nice kata, got no big remarks, well done!

Also, anomalies may appear among the points, which are points located at a distance of one conventional unit from the clusters.

This feels like the wrong spot to convey this information. It's the closest we get to an explicit definition of what a cluster is, and it's burried in an exception. Also, I believe it should be "more than one unit".

Random test generation is causing too much variation in timing. The reference solution occasionally times out, and you'll need to decide whether you want slower implementations to pass or not (e.g. my unoptimized approach passes occasionally).

yes

ranks cannot be changed

Done. Updated the description and removed the necessity for lowercase Latin letters being given consecutively.

1. Everything can be done just because equalityis of a certain type : $\sum_{i=1}^k \lambda_i x_i = n$

   This is not mentioned in the description. (and Katex shoul be used to write the equality)

2. The fact that only consecutively assigned lowercase Latin letters will be given is not relevant here since we only need distinct variables.

Fixed.

Last random tests describe block has wrong name (Should be Large / Big not Small).

Fixed.