Sclerocephalus

That Labyrinthian mesh issue was caused by a bug in the compressed mesh type collision calculations of NifTools which has been corrected in the meantime. If the meshes in question had their collisions built with the latest version of NifTools, there should be no weather issues (and if they have no compressed mesh type collision, it shouldn't matter anyway). From my own experiences with faulty meshes, the CK is much less forgiving than the game. Thus, the lack of crash reports does not necessarily support the integrity of the meshes. It just tells us that few (if any) users of that mod are doing work in the CK. About three years ago, I posted a note on frequent FO4 armor and clothing mesh errors. Any mesh with one of these errors still works in the game but crashes the CK immediately [NB: if there wasn't that killer reply by Insane Plumber which basically says that you must be a complete idiot if you allow for faulty meshes being a problem in your game, there might have been replies by other people and a discussion could have been started. Unfortunately though, with that post in place, anyone posting a reply related to the topic would have had to admit that he's a complete idiot too, so it's not a big suprise that nobody did. Let's just hope that this guy doesn't earn his living as a teacher.] Oldrim behaves in the same way as FO4 in that respect, but SSE appears to be somewhat less forgiving.

Some things to keep in mind: When an actor is placed, the game will try to move it onto the nearest navmesh without saying. If a cell is not loaded, there will be no navmesh either, so placement is expected to be pretty inaccurate. Placing an actor in an unloaded cell via PlaceAtMe won't work if the ref to place the actor at is not peristent. While you can force any ref to be persistent by ticking the respective flag in the CK, having too many peristent refs is not a good idea (they permanently consume memory). Likewise, actors cannot interact with refs in unloaded cells unless they are persistent. Actors that run packages to move from one place to another do still move while they are in unloaded area, but this "movement" is a simulation and no real movement. The game updates them at pretty long time intervals and then moves them some distance forward, so they actually move discontinuously.. Update intervals for actors in loaded cells are much shorter.. Thus, if a package starts running on an actor in an unloaded cell, it has to be expected that there will be some time passing before he actually starts moving (plus, he will not really move until he is on a working navmesh; it's likely therefore that the game will teleport him onto the edge of the loaded area, and only then the package runs normally). It's also worth mentioning that movement through unloaded cells is poorly repeatable. Some years ago, I made a mod for measuring the time needed by an actor to move from one location to another (with a long list of locations covered) in order to find out whether they were repeatable (see https://www.nexusmods.com/skyrim/mods/31241). Unfortunately though, they varied by a very large margin. Of course, reproducibility (which takes hardware and game setup variations on other peoples' PCs into account) is even much worse. Placing an actor may take a considerable real time since the game may need to load his face preset and outfit items. If the actors in question only use vanilla items, this is almost negligible. With mods installed that add more faces (and some mods hold them in very long leveled lists) and/or custom outfit items (with texture resolutions notoriously over the top), the situation can be significantly worse. It's a good idea therefore to spawn actors (even disabled ones) at a far enough distance. Otherwise, they may pop in in plain sight. The more actors are spawned at the same time, the worse this gets. There's a reason why most random encounters start running when the trigger loads, i.e. when the cell with the trigger attaches at the edge of the loaded cell grid.

The original beta had German mod description texts on the muffled mods because I forgot to replace them after I did alter the keywords. It's up to you to decide whether that's important.

3D game objects generally consist of two meshes, one for the visuals (i.e. what you see) and an invisible one for the physics (the so-called collision mesh). The mesh for the visuals has pretty high detail.. The collision mesh however has low detail because the performance that needs to be invested in collision detection increases with the number of faces. A good collision mesh preserves the shape of the visuals mesh as accurately as possible despite having a much lower face count. To accomplish this, tiny wrinkles that may exist in the visuals mesh get flattened out in the collision mesh. Also, accuracy may get reduced on purpose in places that are out of reach to colliding objects. The water tower, for example, does actually not need any elaborate collision at the top because that's out of reach to the player, and there are no high-flying actors around. If collision is noticeably inaccurate (the "thin transparent layer" phenomenon you were referring to) in places that can be directly interacted with, this is plain simply a mesh error and by no means a limitation of contemporary computer games. In fact, you can realize "perfect contact" quite easily; that only depends on the skills of the 3D artist. Now if you look at the tower mesh in the CK and toggle the collision geometry (by pressing F4 while the render window is active, note that you will have to wait a few seconds for the collsion meshes to be displayed), you'll see that the collision at the tower's base is pretty accurate, and also that the first aid box is the object with the bad collision (note the completely unnecessary details on the edges). That box was also not floating but clipping into the tower base instead, in addition to being placed in a position that blatantly derides real-world physics. That's a crystal clear placed reference issue (see screenshot).

The skeletons are static objects, not actors. That is, they will behave just like any other pre-placed rubble piece in the game. They were never intended to clean up.

It's now actually working.

Package records, conditions sub-records: there's a "parameter #3" listed by xEdit in each package of condition data (CTDA): It appears that this parameter may have different meanings - or may be totally unused - depending on what target the condition has been specified to run. If that target is a quest alias (i.e. the "RunOn" line will display "QuestAlias"), parameter #3 is always the ID of that alias on the quest that owns the package (I confirmed this myself by playing around with the aliases).

Which ghouls ? The pre-placed DEAD ones or those that you killed (i.e. the bodies of the pre-placed LIVE actors) We only handle pre-placed DEAD ones because they are never cleaned up otherwise. Pre-placed LIVE actors are clened up by the engine eventually after they have been killed. Though, in a very fresh game with few actors loaded so far, this may take a while. Side note: altering the time scale won't change anything because scripts need real time to process their tasks. This time won't change, so the only thing gained by altering the time scale is a longer game time span for scripts to finish running.

Since we cannot disable refs when there is a risk that this could happen in plain sight of the player, we have to wait until the area unloads. More specifically, until all refs that need to be disabled have unloaded. This process doesn't start running until the player owns the respective workshop. Therefore, it usually takes place the first time the area unloads after the player gained control of the workshop. The Sunshine Tidings cleanup missed two ghoul bodies because they are in cells that do not form part of the workshop location, although they're still within the buildable area (Bug #23763). This has been fixed for UFO4P 2.0.4 but that fix was not retroactive (i.e. they won't be disabled unless you start a new game).

It doesn't but that's not the point. A faulty mesh will corrupt the engine code that handles the mesh and you get a crash. Whether everything else in the loaded area is fine or not doesn't matter.

Once the problem occurs, disable the mole rats, then try to leave the cell.

I'm currently using NifSkope 2.0 Dev 7. Used this to change the user version numbers from Skyrim to FO3 (?), 12/83 -> 11/34., then saved (Note: the file had previously been stripped of the collision, the BSShaderProperty blocks and the BSX flags). Upon trying to reload it, I got this warning: [0] Array Properties invalid Pos 376: failed to load block number 0 (BSFadeNode) previous block was NiHeader Failed to load F:/Skyrim Mod Projects/Nifs for Blender Import/dwermsmcolumnfixed01_raw_01.nif Loaded the stripped file in old NifSkope, changed the user version numbers there and saved. This file loads fine (i.e. without that error) in both old NifSkope and NifSkope 2.0.