Thank you for Uveso!

v239 Update 20 changes, working through the last 6 of Radde's games vs M28 on a particular mapgen map: Adjustments to engineers assisting/repairing an arti/game-ender unit (fixing a bug where sometimes they wouldnt pause when stalling mass, and having them receive new orders periodically rather than only once the experimental is constructed) Refinements to shield cycling logic and associated engineer construction so engineers are more likely to be sheltered by shields when constructing a game-ender, and shield cycling should try and re-align its cycling if disrupted by artillery Torpedo bombers should be better able to target ships that are on the shore-line (so bombs are less likely to hit the beach), and should try and over-kill cruisers if they have spare torp bombers Land units should treat nearby battleships as a threat when assessing whether to attack or retreat Fixed a bug that meant M28 wasn't properly taking into account PD in adjacent zones when deciding if it was safe to tele-snipe Acknowledgements Radde - 6 2v2 replays against v234 M28 on a 'setons-like' mapgen (a series of games on the same mapgen map over a week that has accounted for more than 60 of the changes in v236-239)

Hi, I'd recommend M28 for better mod compatibility. Although in theory M27 should work with unit mods, I'm not actively looking to expand it's compatibility, and in the case of BrewLAN as the FAF vault version is broken (at least the last few times I've checked) it's not something I'd be looking to fix at this stage (although if the FAF vault version does get fixed then I'd be open to investigating the issue you mention of M27 not working at all with the mod enabled)

V2 update - fixed a bug where attacking units wouldnt get new orders when their target was dead. Also updated the wiki with a diagram of how the logic works: https://wiki.faforever.com/en/Development/AI/AI-Modding#overview-of-mini27ais-logic

A new version of the mod, AI-Swarm (Fixed) has been uploaded to the FAF vault. This is based on the latest release of AI-Swarm (v285) by Azraaa, with some fixes by Relent0r so that it works in FAF again.

Updated the Header and the first few posts to be pertinent to the upcoming RCVII. Deleted many posts relating to the last 2 RC's, so... many of the comments by others, map appear out of context, to this upcoming tournament. Will start posting updates and maps on this forum in next few days. Any questions about the upcoming Tournament are welcome below. Thanks for feedback.

@orangerobot Were you playing with any mods? In the Create Game screen, which Game Type did you have selected (FAF, FAF Beta Balance, FAF Develop or Nomads)? Also, please go to client > top left menu > Show logs folder > logs > sort by Date Modified. From there, drag/drop the most recent game log file (should be of the game that you experienced this issue in) into here so those that can help (not me) have that data to look at.

I believe Uveso updated his AI today, Swarm is currently waiting on the developer to update.

I am playing a lot on the dark heart so i can give some feedback, i am not sure though if those observations only apply to this map. I switched from M27 to M28 as ally because he stays in base with com and i avoid those early 5-15min deaths that ruin the game. 1v1 vs M27 is a different story though, as there he will be attacking u with com and it will be hard to defend + snipe him. So as far as M27, if i was doing things i would set a hard time limit as to how long commander is allowed outside the base. Snipes from 10 onwards are to easy and he mostly has no air cover. If you wonder far enough from base, it gives enemies plenty of time to make enough snipe tools even if you start straight for your base right away. I noticed that M27 and RNG, even in minute 30, they will occasionaly venture to middle of map, collect some mass, extractors .. again, i think i would set a hard limit and put commander to the safest place possible. The fact that M27 wonders around area marked with yellow makes me belive he thinks he is safe in the water. I would prefer him moved to the water behind the island, with enough AA on main island to hopefuly prevent any over island torp bomber attempts. M28 is as far as economy goes not far from M27, i even like the fact that he builds factories in remote mass extractor bases sometimes. But with that mass he seems to be less dangerous. But once base is built, you have a very reliable ally at home that will keep his base safe. 2 ideas that i had but i am not sure if its even possible to implement, i would have to look at the files myself. First is use of signals. For instance, even in best scenarios, something can go wrong and commander will maybe follow enemy engineer and wonder far away. It would be cool if you could use 1 of the 3 signals and that would move his com to that location. As an alternative, i see M27 and M28 greet eachother at the start of battle, maybe they could also listen to some of the ally advices, like, move com back to base, help me in my base, hold skies clear. That would simulate a real person ally to some degree. Keep up the good work and if you have any ideas of interesting scenarios i can simulate, let me know. So far i am enjoying those AI's i use more than real players. And my playing has improved a lot more than it did before where in 6v6 i did the always same predictable things and enemy also always was completely predictable and either killed you before you could learn anything or was so bad that it offered no learning experience. ps: i used M27, M28 and RNG in quite a few battles so far and had absolutely no tehnical problems, desync, lagging, etc. I did have issues with core game itself, mostly things like putting bug on patrol in the mid of base, only to see it follow a spy plane back to enemy base and get killed when i look elsewhere. Same with commander on patrol. Should be limited by how far from patrol area u can run after an unit before going back. Probably i am doing something wrong, using the wrong command. [image: 1689103631287-the_dark_heart_preview.jpg]

Well maybe the platoon builder has a template that is only using engineers, or Subcommander are excluded from the assist function. SACUs are excluded from the engineer platoons here: https://github.com/FAForever/fa/blob/deploy/fafdevelop/lua/AI/OpAI/BaseManagerPlatoonThreads.lua#L45 Also SACUs are not allowed to perma assist a factory: https://github.com/FAForever/fa/blob/deploy/fafdevelop/lua/AI/OpAI/BaseManagerPlatoonThreads.lua#L45 These need to be changed or those functions are useless for platoons with SACUs.

There's such a thing as too spicy!

@maudlin27 if you want to know what the AI is actual building and what platoonbuilder is used for it, then you could use my AI mod to debug this. If activated my mod provides the following debug outputs: DEBUG: AI Platoon names will show the name of the actual platoon on every unit. DEBUG: AI BuilderManager will print the actual ordered buildplatoon to the moho debug window ([F9]) [Edit] if you need help with debugging, just post your mod here or on discord and we will look at it.

Good summary. A lot really has to do with both the AI's intentions with naval expansions - and the map makers understanding of how the AI interacts with the markers. Each AI can, and often does, have different parameters for WHEN to expand, and how far to expand. As a result, the experience can vary rather wildly. The AI developers have recently been talking about some logic to better understand what kind of naval investment is necessary, depending upon just how big a map is - and what % of the map is actually water. The idea being that the AI becomes aware that no matter what the markers may indicate - this map just isn't viable for naval work - or it's so significant that it should work much harder on naval.