I'd like to offer my feedback and discuss the AI with the team if possible. I have made several AI related mods for TW:W and recently released a compilation of those that quickly became the most popular AI mod on Steam Workshop: steamcommunity.com/sharedfiles/filedetails/?id=1120180673
The results are very good and the community feedback is excellent so far. Now based on all this work I was of course eager to see what you guys did in TW:W2 and here's what I think of the most critical issues.
First of all I won't be discussing the battle AI here as it is not moddable anyway. What I think of it is that it is very much perfectible but still the best in the series so far. I have no doubt it will be improved further in the next title.- Passive and coward campaign AI
This was one of the most noticeable issue in TW:W1 and yes I said "was" because you guys fixed it in TW:W2, good job! Fix is in the cai_personality_strategic_components table, I have found that a sweet spot was a value of 0.85 for the enemy_strength_modifier (and threat too actually), but I didn't want the AI to be too careless so I use 0.9 instead. I see that by default you now use a value of 0.85 (0.65 for aggressive but hey, they are aggressive ^^). The retreat ratio was good at 0.6-0.7, nice to see you also tweaked this. Anyway, good job there I'm sure the players appreciate it.- AI not assaulting fortified settlements
Another issue was the fact that the AI most of the time waited it out completely instead of attacking with a good chance of success. I found that the value for it in cai_variables was too high for CAI_VARIABLE_FORTIFIED_SETTLEMENT_FORTIFICATION_MULTIPLIER (value of 8). in my mod I use a value of 6 for it and a value of 2 for CAI_VARIABLE_FORTIFIED_SETTLEMENT_FORTIFICATION_MULTIPLIER_SIEGE_ENGINE_DECREASE. That did it nicely. Now I see you also fixed this in TW:W2 with a value of 5 and 1.75, nice job here as well
It'd be nice to see those above changes be applied to TW:W1 as well in a future update, if there is one.- AI character building is completely random
This is also a big issue, not only when you confederate and inherit a useless Lord (or worse, Legendary Lord), but also for the overall difficulty of the game. Battling weak enemy Lords, especially at high level, kills all the challenge. I see that you added the hint USEFUL_ALWAYS for several key skills and mounts in TW:W2 and that's also a very nice change in comparison to the first game where everything was random. Maybe you should add the spells as USEFUL_ALWAYS too, I saw people get a Morathi or a Teclis by confederation who didn't have any spell! You also created the much needed USELESS_ALWAYS hint. Could it be created for TW:W1 as well?
Now let's talk about the two major things IMO, one very good and one... sorry guys... still very bad.- AI settlements building
That's the good one. When I was working on my mod I quickly realized that good AI armies (by good I mean both fun and challenging) needed good province building. I managed to tell the AI WHAT to build by tweaking the score_inclusive in cai_construction_system_building_values and of course by adding many synergies in cai_construction_system_synergies. Still it was not possible to tell the AI WHERE to build. I see that now you added the scope of the synergy and it is now possible to tell the AI WHERE to build thanks to this. Well guys, thank you very much for this and I hope you will use it for Mortal Empires (as of now it is only used for the New World core races). Here also, could the scope be created for TW:W1 as well?- AI recruitment logic
Now the bad... Sorry but the AI recruitment is still very bad. Better than what it was in TW:W1 at release of course but still, this is a major issue, the most important I'd say and unfortunately it didn't get the attention it deserves. I'll talk about the cdir_military_generator_template_ratio and cdir_military_generator_unit_qualities tables.
My most important question is: do the templates work in campaign or is it only for the random army generator in custom battle?
This is not a stupid question as I found out that tweaking the templates had next to no effect in campaign, while tweaking the unit quality values had an immediate effect on what the AI recruits. When you see the AI armies in campaign you'll see that almost all of them are absolutely not close to what they should be if they followed the templates. And of course I'm talking armies that were recruited while the AI had access to all unit types mentioned in the templates. That's the same in TW:W2. I'm sure many people here have noticed the Dragons/Phoenix spam among other things. We all remember the Norsca chariot spam even though the AI had access to the other units. The chariots value was so high that the AI only recruited those despite the template ratio for this unit being at something like 15-20% per army. I fixed this in my mod just by reducing the unit quality value and not even by changing the ratio. This is just one example among many. In the templates, those units should be limited by the ratios but that's not the case, their unit quality values being much higher than the other units in the roster, the AI tend to spam them. Tweaking the unit quality values immediately fix that issue with the AI spamming the highest one no matter what its unit type is. Tweaking the templates has no effect on it. Does the unit quality value takes precedence over the template ratios? If that is the case you can see how this work against unit variety in AI armies.
In my mod I achieved a great result with AI army compositions by tweaking the unit quality values and not the templates. The values you use (unit cost multiplied by the unit tier) is logical but lead to the AI ignoring the template ratios. I basically reduced the gap between tiers and even reduced a lot the value of the highest tier units and it works quite well. I'd really like your opinion on this and I wanted to check with you to see if you're aware of this (template ratios not taking into account).
That's all for now. @CA_Whelan
if you guys could forward this to the team please? Thanks.