Does anyone have feedback on the new "Aggressiveness = 1"-behavior in 8.5.5?

[Xilmi]

I think it is the closest I have gotten to mimicking the behavior of a human who is trying to clear the map.

I had something similar in the past but there are some distinctions that should make the current one better.

Notably a distinction in behavior depending on whether a unit thinks it was spotted or not.

A unit who doesn't think they were spotted and also doesn't know about any enemies' location will first peek and then move to the best cover that is closer to where they assume the enemy than their current location.

However, if they think they are spotted, they will go just go for the best cover overall.

The difference this makes when it comes to my previous attempt to player-like gameplay is that you can't exploit it in the same way with camping as they are no longer forced to move further ahead when they were spotted.

Overall it performs a little worse than aggression 0 as that one allows them to create much better ambushes on their own but it's a massive step up from the previous aggression 1, that could easily turn into a shooting-gallery.

[Xilmi]

By using a combination of nearness and cover-value for scoring tiles it often would get as close as it could, so not really staying in good cover.
Another issue is that when it doesn't have any cover around it would just dash forward. Dashing forward is good for melee or short-range-units but ranged units probably should inch forward more slowly and preserve more TUs for reaction-fire.

Also there's cases where the units looked at walls at the end of their turn.

[Xilmi]

It's frustrating. I implemented it to work exactly as I intended it to work and when testing it simply turned out to be just bad.
Due to mutual surprise, it doesn't even really matter all that much if you preserve your TUs. You get spotted and killed by snipers/grenadiers from outside your vision anyways.

I also experimented with a completely different approach that first positions all units outside the assumed danger-zone and then rushes everyone forward. While in theory slightly more effective than rushing forward from the get-go, it also allows the enemy to prepare better.

Overall, being the aggressor is just dramatically lowering the chances of a favorable outcome. It only works when you have a massive numerical advantage. There also isn't really any in-between tactic that's favorable to outright rushing. Or at least it only works when the enemy also does an in-between-tactic by tricking the enemy into thinking they have to search for you while you are still remaining hidden. But as soon as the point comes where you'd have to leave your cover, it's very easy to exploit.

Look for enemies out in the open and hide behind cover again if you don't find any simply is by far the most effective tactic available.

There isn't really any room for "intermediate" approaches.

[Abyss]

I didn't test yet, because I really want to start testing while making modification of RUL file for, say, XCF mod with, hopefully, non-whole aggressiveness/behavior possibilities. But some things might be insightful:
- There is lots of countermeasures against campers. Smoke grenades. Invisibility. Preventive rocketing/explosions on AOE (no-visibility/intuition). Tanking with heavily-armored unit. Speed.
While many of these things are still a challenge for AI to get into use, speed and tanking are matter of calculations.
I don't really remember the reaction fire chance formula, but I believe it depends on difference in reactions and remaining TU's between moving unit and observer. It's quite usual for middle-stage modded game that player-controlled units have 200 TU's and good reactions/bravery. These units will trigger much less reaction shots than ordinary 80 TU unit.
Tanking is different matter. If BAI could make such decision for better outcome without cheating, it would be nice too.
- The outcome is very sensitive to game balance, and it should be taken into account. There's limited profit to train BAI on fast-battles, because they represent only 1% of lategame scenarios, while major of most intense and interesting battles happen at early/middle game stage.
Balance is constantly shifting during the game: early-game missions yet appear (if mission source is not squelched), while new challenges appear. And those can be hard or easy.
 Imagine a mission vs one single naked psycho with nuclear rocket launcher. One chance mission, he can fuckup the whole squad on turn one, while others - he will remain helpless in CQC, if player manages to go unnoticed. It's all about thriller
 - As player, I will keep saying it again and again: there is 500-1000 missions to play. That should be, at least, fun. 

     

[Xilmi]

It doesn't work like in 8.5.5 anyways and I figured it was pretty bad and more like combining the weaknesses and not strengths of other approaches.

At some point there's only so much the AI can do. If the player can start leveling entire building-blocks with everyone inside without even needing to establish any kind of vision, it doesn't matter much what the AI does.

Note that camping isn't so much about relying on reaction-shots. I know this is a bad approach against a player who knows what they are doing. They will still try their best spotting the player on their turn but they won't really advance towards where they think player-units are and instead stay close to where they spawned.

Like with none of the previously existing levels you really had to explore the 3rd floor of a UFO because they would all have moved to the 1st one by the time you got there. That's what the new lowest level is meant for.