[SOURCEMOD] Brutal-OXCE 7.12.1

[Xilmi]

I checked it on GitHub and README seems to be borrowed from OXCE. Do you have your own with all features and configurations and defaults explained?

The first post in this thread explains all the new options. But you are right, I should probably overwrite the readme, which up until this point is simply the unmodified one from OXCE, which I have forked from.

Also it seems to be a combination of AI with QoL with some other modded elements. QoL elements are good to have but is it possible to separate AI from other modding? Or, at least, to turn all other modifications off by default leaving AI changes only? Thank you.

Do you have any concrete examples for QoL-improvements that currently are enabled by default but you want to be disabled? And can you give a comprehensible reasoning as for why they should be disabled?

I'm not a fan of generalization and would rather look at each topic individually to make the best decision for it.

[Alpha Centauri Bear]

Sure thing. Thank you for responding. This is a revolutionary mod generally.

The problem with such mingling all kind of things together: AI, mods, QoL is that you are building mod for yourself. That is good. Many people will play it because they like the same style. However, many people will also not just because they don't like a thing or two you mingled into the mix. Not to criticize, but just trying to understand what are your default settings to decide for myself.

Here are few examples I grabbed from first post.

Enhanced dogfight behavior
Realistic accuracy and cover system
One-click grenade priming
Fog of war
Smart ctrl-click and auto-equip

[Alpha Centauri Bear]

Another thing I experienced while developing AI for other games is that players generally do not care about AI fine tuning. In their perception the opponent is just smarter or dumber. And this is what probably you want them to tune up or make it dynamical by difficulty level or something like that.

I hardly can imagine a player that turn off a specific element of AI behavior (like aliens using better cover) just because it does not match their player style (?) .

In other words, I would recommend to set one singe dial "AI strength" and put all details under the hood. This way your implementation is independent on all these parameters. You will be AI designer and not just provide service methods for players to tune it up themselves.

[Juku121]

I hardly can imagine a player that turn off a specific element of AI behavior (like aliens using better cover) just because it does not match their player style (?) .

I've never played a game with customisable difficulty without my final difficulty being called 'Custom' or some equivalent. So I'd say your imagination needs an upgrade.

In other words, I would recommend to set one singe dial "AI strength" and put all details under the hood. This way your implementation is independent on all these parameters. You will be AI designer and not just provide service methods for players to tune it up themselves.

While this is something Xilmi has been musing about on their own, I for one would appreciate if the fine-tuning still remained an option. If not an in-game option, then at least a set of ruleset or config variables, hidden from the eyes of the 'filthy casuals'.



@jnarical: I get your ideas about tactical approaches to clawing back the 'lost' accuracy, but there are two problems with that. First, AI doesn't really have the same toolset as the player does, unless Xilmi codes in a separate branch for 'real accuracy'. They won't show up in jungle maps with flamethrowers, never mind actually using them against the foliage. They won't kneel unless a mod makes them able to kneel. Etc.

Second, "it's fine if the base chance to hit is minuscule, use tactics" is the same line of thought that led to JA2's 'new chance to hit'. And even with a bazillion levers to tweak it and people sharing their setups, it never really clicked for most. I really like this concept, but I'm afraid it'll go the way of the dodo nCtH if the gameplay impact is tailored to just one individual's preferences.

[jnarical]

Second, "it's fine if the base chance to hit is minuscule, use tactics" is the same line of thought that led to JA2's 'new chance to hit'. And even with a bazillion levers to tweak it and people sharing their setups, it never really clicked for most. I really like this concept, but I'm afraid it'll go the way of the dodo nCtH if the gameplay impact is tailored to just one individual's preferences.

Could you please be more specific? What’s wrong and what you think would be better to do, to fix that?

I get the idea “we don’t like hard caps” but I don’t see any good alternatives. That doesn’t mean there isn’t any, just that I need some guidance )

[Juku121]

Essentially, what's wrong (IMO, I don't claim any authority here) is that the current system is extensively playtested and works at least somewhat, and yours is a radical departure that at least at a cursory glance increases time spent on already excessive mission grind in an 'unfun' way (who likes to miss a lot? or even be missed a lot?). Well, unless all you play is short vanilla campaigns or vanilla-ish mods, which I no longer have the stomach for.

But, in general, my opinion is that excessively low hit chances (something like sub-30% or so for most shots) do not make for interesting gameplay.

I tried my damnedest to tweak the JA2 nCtH to work for myself, and in the end it kinda worked for me, but once I looked back at the vanilla JA2 accuracy model and asked myself "is this better?", it was not. Maybe it was not worse, but it was a lot of effort to stand still at best. Yeah, it was more 'realistic'. It was not more fun.

As to the current implementation, I'd need to do some more testing to speak with confidence (and I just lost my filters, again, so no X-Com until I restart my machine, which might take a little while ), but what I can definitely say is that:

1) Having a "model of low accuracy + tactics" means that modders lose some of their current control over accuracy. That is, while the game no longer outright lies and misleads, you now have to jump through extra hoops to match your vision and in-game reality.

1b) Someone has to give the 'tactics' to the AI, tactics that are different from the vanilla model.

2) I don't mind caps as such. I mind drastic caps like accuracy never going above 95% except for the rarest of circumstances, or not under 5%. I might be biased here, because I've grown to dislike d20 games a lot due to similar caps. It fits well into a storytelling game where interesting events, good or bad, must not be too far apart. It's absolutely annoying in a simulationist wargame or tactics game.

1%/99% caps, sure, go ahead.



TLDR, I agree with Abyss that there should be a tweakable 'cover factor' for accuracy, at least. Maybe not for the player, but certainly for the modder to experiment with.

Auto-shots should only have the first shot use the accuracy-as-visible area model, and the the other shots should just scatter around the first shot in a cone or, better yet, rising/falling line. Ditto for shotgun pellets. And a horisontal line for auto-shot waypoint guns.


And, in general, I'd actually like it more if the cone was fixed, and thus accuracy was only ever a stat at a certain range (max range?, half max range?) and all other accuracies were actually calculated from that, using a fixed cone. Possibly with some extra input from all the dropoff values, but without them directly translating into accuracy percentages. This might not be what you're trying to accomplish here, though, and you certainly have no obligation to listen to me.

[jnarical]

[mention]Juku121 [/mention] as I’ve said earlier, upper cap isn’t actually the cap. There’s gradual increase of accuracy up to 100%, based on distance in voxels between shooter’s weapon and nearest point on target’s cylinder. For distance = 0 there could be 100% accuracy, but usually there’s a gap between two units, as they stay in the centres of their tiles, so distance never becomes 0, and accuracy never becomes 100%. For some units and relative positioning, the distance is minimal and that gives maximum possible accuracy.

For minimal cap situation differs but it’s still possible to get 1% visibility, which is minimum possible value, other then 0% which means target doesn’t have line of sight. Most of the time, minimal cap is 5%

I could change that, in theory, but can’t afford spare time at the moment

[Alpha Centauri Bear]

Another newbie question.
Do I understand correctly, you don't care about any compatibility with OXCE? If so, how would one install your mod and then install some other regular OXCE mods on top? Is it even possible?

[jnarical]

Another newbie question.
Do I understand correctly, you don't care about any compatibility with OXCE? If so, how would one install your mod and then install some other regular OXCE mods on top? Is it even possible?

Question to me or Xilmi ?

[Xilmi]

1b) Someone has to give the 'tactics' to the AI, tactics that are different from the vanilla model.

You are right. It should actually be pretty easy. For target-selection incase several targets are attackable from one position that's already done but not for attack-spot considerations itself.

When realistic accuracy is enabled the checkVoxelExposure is run in AI code in order to avoid false-positives. When I do that anyways, I can also factor in that value into the score for the tile to attack from. So if moving a tile further drastically increases exposure, it should then notice that and do so.

So thanks for the hint!

[Xilmi]

Another newbie question.
Do I understand correctly, you don't care about any compatibility with OXCE? If so, how would one install your mod and then install some other regular OXCE mods on top? Is it even possible?

Hu? What gave you this idea?

A lot of effort was put into achieving compatibility with various mods. Handling immobile units like turrets, which don't exist in vanilla, handling neutrals that shouldn't be attackable like damsels, taking short-ranged- but non-melee-weapons into account in different scenarios. Oh and of couse the weird hand-to-hand-combat-mechanics, where you miss all your shots in melee-range because the enemy supposedly grabs the barrel of your gun and points it somewhere away from them. That's also supported.

I wouldn't go as far as to claim that Brutal-AI is fully functionally compatible with absolutely everything that regular OXCE is compatible with. But that was a goal and if someone reports a bug for something that works with regular AI but causes an unintended bug with brutal AI, I'd try to fix it.

The idea is that you can play every mod that is compatible with OXCE on Brutal-OXCE and with enabled Brutal-AI too.

Whether the mods are then too difficult to be actually winable is another story. I'd say tone down difficulty-setting and choose some of the easier brutal-AI-settings. Especially the recently introduced intelligence-option.

Edit: And if this was about "Realistic Accuracy". It's just an optional feature, which is disabled by default. If you don't like how it interacts with certain mods, simply don't use it.

[Alpha Centauri Bear]

Thank you for confirming.
In this case where I can find a list of configuration parameters with defaults? I unpacked the zip but there is no readme there too.

[Alpha Centauri Bear]

OXCE launch file is OpenXcomEx.exe
Yours is OpenXcom.exe

Should I launch yours?

[Alpha Centauri Bear]

What is "Realistic accuracy and cover system"?

I understand vanilla already accounts for cover in the way that if part of the target is covered and shot hits the cover it does not hit the target. Sure this can be calculated and displayed to user for convenience but what exactly is changed?

[zee_ra]

What is "Realistic accuracy and cover system"?

I understand vanilla already accounts for cover in the way that if part of the target is covered and shot hits the cover it does not hit the target. Sure this can be calculated and displayed to user for convenience but what exactly is changed?

Actually, that is a rather interesting question.  I think, vanilla already accounts for voxel shapes on the way.  Is the model different in this release of the engine?  If so, how it is different?