Considerations on the Genko Campaign (Japanese side)

[GShock112]

The mongols seem quite strong and I've never faced their units before so I'll have to learn some new tactics... After an attempted engagement with a strategic retreat, I managed to have a skirmish in Kyoto with 100pts disadvantage (a severe numerical disadvantage) but I won it. It goes without saying if I increase the difficulty level the enemy will retreat more, combine more armies and then fight with even superior odds. It's more cat & mouse and less combat with realistic odds and I don't like it.

A few things I noticed here:

1) The mongols did not make a single move until I tried to reach the center of the battlefield. Half of the 20 turns could have passed with nothing happening... but in this case I was attacking Kyoto. The AI as usual came to meet me instead of camping and shooting... this is very bad considering most mongol units are missile-equipped.
2) 2 enemy skirmishers tried the usual flanking FAR manouver (if that's the plan they should start it immediately). It started when they all moved, not earlier as appropriate, and it wasn't carried out because they fell back to try to flank me when it was too late. The movement system costs a lot of points, this kind of flanking apparently can't be pursued even by cavalry skirmishers if the AI doesn't start it much sooner and some sort of safety must be adopted so that this manouver is treated properly otherwise it's just basically taking units away from battle and instead of being advantageous it turns into a disadvantage for the AI.
3) A retreated skirmisher in disrupted state came back to firing range... and it retreated at once. Either retreat and stay far or don't retreat, period.
4) With superior number of units, the AI should focus more on flanking.
5) I am absolutely sure I made the right move when moving my far army from Echigo to Kiso... however the army was moved to Kai. The "borders" of these provinces should be checked for glitches.

This is another battle the AI would have won without these 3 mistakes which I've come to find familiar by now... even with those, the 11% 44% with which I won do not give justice to the real battle because half of the enemy units were not committed and because I was lucky in pursuits catching the AI 2nd line on their flank.

What happens to the wounded (how many in % are recovered and how many end up dead) and to the missing/deserters? Is this connected to the ownership of the province where the battle took place at the end of battle? (It should)

[rbodleyscott]

2) 2 enemy skirmishers tried the usual flanking FAR manouver (if that's the plan they should start it immediately). It started when they all moved, not earlier as appropriate, and it wasn't carried out because they fell back to try to flank me when it was too late. The movement system costs a lot of points, this kind of flanking apparently can't be pursued even by cavalry skirmishers if the AI doesn't start it much sooner and some sort of safety must be adopted so that this manouver is treated properly otherwise it's just basically taking units away from battle and instead of being advantageous it turns into a disadvantage for the AI.

This issue will be addressed in the next patch.

5) I am absolutely sure I made the right move when moving my far army from Echigo to Kiso... however the army was moved to Kai. The "borders" of these provinces should be checked for glitches.

I have just tested this repeatedly and it seems to be working correctly.

What happens to the wounded (how many in % are recovered and how many end up dead) and to the missing/deserters? Is this connected to the ownership of the province where the battle took place at the end of battle? (It should)

The reported wounded already takes into account those that are recovered. They are simply not reported. The % that are recovered depends on who won the battle - obviously the winners can tend their wounded whereas the losers mostly have to leave them behind.

[GShock112]

It is very important you find a way to teach the AI that when they are attacking a province they must attack aggressive with the melee units and when they are defending a province they must not come forward to meet the player in center-field but use missiles as the player advances. Technically I can advance my units from the initial deployment for 2 turns and the AI won't move... but then it surges forward, not aggressive enough and when the 1st line is routed, the second line is not ready. If that second line was in the flanks or farther away (in reserve, so as to not be involved in the morale loss), it would be completely different (they would catch my pursuers).

I'm sure you guys have read enough from my 2 Tenka Fubu campaigns to see that the AI tactical mistakes are recurring... and yet... the AI is doing a very credible job considering how complex the system is... in fact... even when I win I can say it could have gone the either way. You really did good as it must have been super hard to program it.

[GShock112]

After a few attempts we did join battle finally and I won... an easy victory (43% - 6%).
This is the ending situation.



The mongols regrouped combining all their armies but could not engage, I combined myself and when I attacked we joined a HUGE battle with excess of 25.000 men per side in Kyoto and both sides in supply shortage (2500+ pts of army values each). This was easily the greatest of all the battles I ever played with this game so far and here's how it looks like on a very complex map.



What is particularly relevant according to me is the difference between how I am organized and compact and how the AI is disorganized when advancing to contact (which means it will come slow and be bombarded heavily). The screenshot is taken BEFORE the AI moves so that's the situation of their deployment at the beginning of their turn and the ending of my previous one. I am organized, they have half the army out of position and half the army in complete disorganization.

Again, it looks like I am defending, in fact, I am in the woods mostly... but it's me attacking Kyoto... the AI should be waiting and instead... it comes to me in the open. That's really a top priority fix: tell the AI what's going on in the campaign must be reflected on the battlefield (if I draw --> I retreat from Kyoto ---> I lose men, so the AI must wait for me to go to them and not the other way around).

I essentially was about to win by 1% when one of my units was routed but luckily I made it up and still met the 1st requirement 2 turns later.

Given the overwhelming number of units in this battle, almost all my pursuers were caught in the flank/rear and I could only save 2 exploiting the gaps.
I won the center with this major push while I lost my right flank but the AI did not converge to the center... on the contrary, it focused on eliminating my last 2 isolated units on the right instead of flanking me to stop my central push.

I did converge instead from the center to support my left flank in serious distress won a moment before I was about to lose that flank too.

So you see... if the AI had had just a bit more time to develop its plan and had not advanced in disorganized fashion, this battle that looked like an easy victory for me, just like all the others could have ended very very bad. It's more like they don't attack all together but each individual unit has a plan of its own... which doesn't work, obviously on the battlefield where the team effort is paramount.

This is the situation now.



I can't believe how easy it feels I was winning until I lost the right flank. If the AI had converged to the center this would have really been a catastrophe.

[kongxinga]

Now I hear what GShock is saying, and I agree the AI could do a better job. On the other hand, I understand that writing good AI, whether classical game AI or the newer data driven ones is VERY HARD. So most game programmers don't have the resources, time , know how or ability to write good AIs apart from scripting somewhat more intelligently in some situations. It's basically the effort of making a game, all over again to create a good AI. And wargames are complex beasts that most people can't understand. I think grognards may underappreciate how select a group they are.

That said, is AI behavior very moddable? I do recall there were some tweakable settings in Pike and shot that I believe would apply here as well that governs whether the AI holds or attacks etc.

[GShock112]

Well... considering how complex this thing is, by no mean easier than chess itself (with the added random factor of dice) I think the AI is doing a remarkable job and while its flaws have been identified, fixing them may take some time. To some extent things can be bettered but there exists no really working AI... I mean the real concept of AI is a generalized "library" of moves but not a thinking program.

To begin with, you can see in the battle screenshot that the AI army is in a state of total disorganization after multiple turns. The biggest problem here is that when hand to hand is joined those units in second line run the risk of being caught on the flank or rear by one of my pursuer (and it happens all the time). So, aside from not being helpful they also trigger a cascade morale loss when they are routed in turn. This movement scheme is the first thing to solve as when the army marches compact it's very hard to find a weak spot... and if there's one it must be created, it will take time, and be done at risk of failure.

What can be done NEXT, is to start from the campaign situation and set 2 major triggers:

1) If the AI is defending a province it must dig in on defensive positions: the player has to attack and rout the AI otherwise he will have to withdraw if he doesn't succeed (and that means he takes attrition losses aside from the losses the player will take anyway when trying to beat the AI, whatever the result will be).
2) If the AI is attacking a province it must attack the player.

In both cases there have to be a few more triggers:

3) If the AI has numerical advantage in melee it must engage all units at the same time and use part of the reserves to flank the player. The other part must be kept ready to fill the gaps of the first-line.
4) if the AI is even in melee numbers it must be cautious.
5) if the AI is at numerical disadvantage it must hold good terrain.

There have to be subtriggers:

6) If the AI has more missiles than the player it must not rush to the melee and if it is inferior in missile power it must rush as fast as possible to the melee. Cavalry must absolutely keep the missiles busy.

I could add 10 more triggers and subtriggers in priority order but these are the most valuable for the best gaming experience.

I can't stress enough the importance of saying that despite these major flaws and despite them being omnipresent, each battle is a new experience and it takes a few rounds without doing mistakes to beat the AI at the Daimyo level. That doesn't always happen. The unpredictable motion of routers and pursuers is a double edged sword.

[GShock112]

And we managed to get a good fight in Niniwa, though the battered mongol army couldn't stand against mine, even with the help of a huge auxilliary force in support.

I am on the offensive, right... then here's the AI coming to me, dug in on higher ground just behind the woods.
I highlighted the front line (look at my lined archers and imagine what happens when the AI comes out of the woods... look at the mini-map and how the AI is advancing compared to how I am deployed), picture how different this battle could have been if I had to be the one to march through those woods (as proper). Result is 40%-11% but it was 40%-0% just at the previous melee phase.



The battle log and the casualties are unforgiving...