"Discretion is the better part of valor."
A lot of stuff is still janky or under development or imbalanced, but it would really, really help if the AI didn't suffer absolute stackwipes constantly. Up front: I'm aware the AI does sometimes break away from bad autocalc fights occasionally. My position is that they should do it almost every time the battle turns against them and it should be difficult to annihilate a full army and capture a half-dozen or more lords in the process without trapping them or some other special circumstance.
First things first though: manual battles. There is stuff in the DLLs about organized retreats and withdrawals of individual formations, great. But I've never seen it actually happen in a battle. Not even one time. I'm not great at reading and understanding code so I can't say whether it is bugged, an oversight or simply unimplemented. I can only speak to having not once seen a lord bail on a fight in the way a player can (properly): by ordering his units to retreat off the map. Yes, sometimes their morale breaks and they rout off the map. But in a deliberate manner? Not once.
The AI shouldn't sit up there watching the powerbar tilt against them and think, "Nah, I'll just ride it out, see what happens, any given Sunday, etc." -- the way battles play out in Bannerlord isn't like real life where you can have wild swings in fortune; it is much, much closer to Lancasterian n-square law (tl;dr a tactical advantage that snowballs so you kill everyone against you for few losses). The only time I've seen a battle actually turn against me was when my troops were killing dudes (who outnumbered them like eight to one at the start) just fine until their arrows gave out.
The majority of players -- and I'm judging by streams and YouTube videos not just my own play -- understand this intuitively even if they don't necessarily retreat for whatever reason. But the AI acts like 600 vs. 800 becoming 500 vs. 750 is still a winnable fight, despite it being obviously not to anyone with a brain and eyes to see the powerbar. (As an aside, the powerbar should represent total power, including reinforcements, unless there is some way to stop reinforcements from spawning.) A fight especially isn't winnable in the case when the superior force has a lot of ranged power (Khuzaits and Vlandians, basically) that can kill without necessarily being endangered.
That last bit about ranged power only applies to manual battles but autocalc has its own issue: hundreds of dice rolls means everything winds up just about average, every time. Just to illustrate what I'm saying: if I roll my 4d6 vs 5d6, I'll have a decent chance of winning. In fact, no one should bat an eye if I win with 4d6 against 5d6, two times out of three. It isn't the most likely outcome but common enough that there is no issue.
However, as the number of dice rolls goes higher, the chances of an upset become lower -- much, much lower.
NOTE: Generic d6 dice rolls, not BL's autocalc being simulated.
Presently, autocalc does a lot of rolls -- I'm not sure exactly how many, but at least one per attacker, per round. Numerically superior forces get an advantage in first attack, so they blow away a portion of the defending side first. In the end, things still usually snowball to a finish. And like I said to start with, occasionally the AI does bail out when its bad. But way too often it either doesn't even try to escape intact or it tries and fails because there are no mechanics to support a fleeing army. Or maybe there are but they aren't strong enough to matter.
This is part of why campaign-level snowballing happens: minor advantages add up into big wins, big wins add up into more advantages in the next fight and so on. It also influences things like the campaign balancing, because the devs have to weigh the fact that the AI gets stackwiped repeatedly. So they get an XP bonus, they get an extra two recruitment slots, they get to start with more parties, more favorable exchange rates on loot, etc.
tl;dr: It would help with the snowballing issue if the AI would bail before it became a truly egregious stomping and there were fewer dice rolls in autocalc.
A lot of stuff is still janky or under development or imbalanced, but it would really, really help if the AI didn't suffer absolute stackwipes constantly. Up front: I'm aware the AI does sometimes break away from bad autocalc fights occasionally. My position is that they should do it almost every time the battle turns against them and it should be difficult to annihilate a full army and capture a half-dozen or more lords in the process without trapping them or some other special circumstance.
First things first though: manual battles. There is stuff in the DLLs about organized retreats and withdrawals of individual formations, great. But I've never seen it actually happen in a battle. Not even one time. I'm not great at reading and understanding code so I can't say whether it is bugged, an oversight or simply unimplemented. I can only speak to having not once seen a lord bail on a fight in the way a player can (properly): by ordering his units to retreat off the map. Yes, sometimes their morale breaks and they rout off the map. But in a deliberate manner? Not once.
The AI shouldn't sit up there watching the powerbar tilt against them and think, "Nah, I'll just ride it out, see what happens, any given Sunday, etc." -- the way battles play out in Bannerlord isn't like real life where you can have wild swings in fortune; it is much, much closer to Lancasterian n-square law (tl;dr a tactical advantage that snowballs so you kill everyone against you for few losses). The only time I've seen a battle actually turn against me was when my troops were killing dudes (who outnumbered them like eight to one at the start) just fine until their arrows gave out.
The majority of players -- and I'm judging by streams and YouTube videos not just my own play -- understand this intuitively even if they don't necessarily retreat for whatever reason. But the AI acts like 600 vs. 800 becoming 500 vs. 750 is still a winnable fight, despite it being obviously not to anyone with a brain and eyes to see the powerbar. (As an aside, the powerbar should represent total power, including reinforcements, unless there is some way to stop reinforcements from spawning.) A fight especially isn't winnable in the case when the superior force has a lot of ranged power (Khuzaits and Vlandians, basically) that can kill without necessarily being endangered.
That last bit about ranged power only applies to manual battles but autocalc has its own issue: hundreds of dice rolls means everything winds up just about average, every time. Just to illustrate what I'm saying: if I roll my 4d6 vs 5d6, I'll have a decent chance of winning. In fact, no one should bat an eye if I win with 4d6 against 5d6, two times out of three. It isn't the most likely outcome but common enough that there is no issue.
However, as the number of dice rolls goes higher, the chances of an upset become lower -- much, much lower.
NOTE: Generic d6 dice rolls, not BL's autocalc being simulated.
Presently, autocalc does a lot of rolls -- I'm not sure exactly how many, but at least one per attacker, per round. Numerically superior forces get an advantage in first attack, so they blow away a portion of the defending side first. In the end, things still usually snowball to a finish. And like I said to start with, occasionally the AI does bail out when its bad. But way too often it either doesn't even try to escape intact or it tries and fails because there are no mechanics to support a fleeing army. Or maybe there are but they aren't strong enough to matter.
This is part of why campaign-level snowballing happens: minor advantages add up into big wins, big wins add up into more advantages in the next fight and so on. It also influences things like the campaign balancing, because the devs have to weigh the fact that the AI gets stackwiped repeatedly. So they get an XP bonus, they get an extra two recruitment slots, they get to start with more parties, more favorable exchange rates on loot, etc.
tl;dr: It would help with the snowballing issue if the AI would bail before it became a truly egregious stomping and there were fewer dice rolls in autocalc.
Last edited:
