Oh! Thanks! Sorry to hear - it used to work for me so far. Will look into this tonight.would you have more details , e.g. the xml that threw this error (byPM if you wsnt)

Thanks for making this @Grautvornix! I really like the wall drawing. I was getting some errors though. Maybe this is tied to that, but the walls didn't appear in atlas /Users/me/Downloads/mercator-0.9.0/atlas_maps/model_picture.py:320: RuntimeWarning: invalid value encountered in matmul normals = piece.normals @ view.T  

New tests on the latest iteration. The bouncing elephants bug is now much harder to reproduce. When trying to gather them at the same point, the issue is much less pronounced than before:   However, I was able to reproduce it in the following situations:     Walking around corners: I think the bouncing effect is triggered when two or more units become blocked by objects or other units of similar strength. You can see in one part of the video that a single elephant gets blocked, but nothing happens because it has enough space to bounce. Later in the video, however, more than one elephant gets blocked from all sides, and that's when the bouncing starts. Yesterday we talked a bit about adding some kind of diminishing return calculation to reduce the bouncing effect as it keeps repeating. This could be a relatively low-cost solution in terms of performance that might address the issue.   Moving through narrow paths: This is perhaps the most important issue to address. At some point, they should be able to "agree" on who gets through first. Common sense tells the user to destroy the house first before trying to pass, and it's perfectly valid for a successful attack to depend on the player's ability to move their units through narrow paths or turtle bases. However, it is still noticeable that more than half of the elephants get stuck in a space they should be able to pass through relatively easily:   It makes sense that they don't turn as quickly as they do in vanilla and that, as a consequence, their attacks and movement take longer to execute. After all, we're talking about a huge animal. Somewhat slower and, in a way, "clumsy" movement is to be expected. Beyond continuing to tune the elephants' physical values, I think it would be worth considering some degree of overlapping. Not as much as what currently exists in vanilla, but enough to help reduce some of the clumsiness and to emulate a slight "compression" of their bodies when they are packed together. After all, they are organic bodies with some degree of flexibility, not rigid objects like tanks. Could we also try making their hitbox oval rather than square? This might help reduce collisions between them even further.   This last comment is outside the scope of this PR, but I think it's worth mentioning: This testing also made me wonder whether it's desirable for elephants to prioritize attacking nearby units over buildings. In the video example, I send the elephants to attack the fortress directly. Then I press H to make them attack freely. The elephants prioritize attacking the units, but the ones in the back can't reach them and end up bouncing around without dealing any damage. Elephants are perhaps the only "hybrid" unit in the game, in the sense that, although they are not classified as siege units, they can still destroy buildings quite effectively. Perhaps we should consider giving them a smaller target acquisition range, so that in situations like this, they choose to attack buildings instead.

One could vibe code a tool which automates that, behind replay pallas the api is public  

Would not reduce the impact of Mercator to his famous projection. As I wrote, he did other things as well (producing an Atlas). Regarding spherical maps, this is specifically a question to the game mechanics, i.e. a unit leaving through one side must reappear on the opposite side.  this might have quite some consequences on pathfinding.