Auto Resolving Sieges

[Sentinel]

Auto Resolving Sieges
In MTW and RTW auto resolving battles was usually a very costly way to fight a battle. It was rare to find one that you could not fight yourself with less losses and more kills.
The way these battles are calculated seems to have been changed in M2TW to give less of an advantage to the AI faction than previously.
This is particularly noticeable during a siege where often auto resolving the siege will give far fewer losses that even the most skilled human player could ever achieve no matter how many times they tried.
This is particularly the case, in the early game, before the availability of better siege engines. A larger attacking force of low quality units would give auto resolved losses for the whole siege lower than the loss experienced by a manual attack just pushing the ram to the gates.
Since I now found myself using the auto resolve more often (arguably too often), I became curious as to which factors affected the calculation. So I set up the following test.

All the testing below is based on my England VH/VH campaign.

Two similar settlements were selected and gifted to the Turks, who garrisoned each of them with 8 units of Mercenary Pavise Crossbowmen.

Castle – Magdeburg - Population 15,086 – Citadel, paved road, crop rotation.
Town - Breslau - Population 16,175 – Large stonewall, paved road, crop rotation, town hall, brothel.

I positioned a number of spies and full stack armies between the two settlements so that both were within striking distance of the any selected force in the same turn. The units used were -
-----------------------------Attack----Defence--Skill-----Shield----Total
Armoured Swordsmen-------14------9---------8--------6--------23
Feudal Knights--------------11------8--------5--------5--------17
Retinue Longbowmen-------12/8-----7--------7--------3--------17
Town Militia-----------------5-------0--------1--------6--------7
Hobilars---------------------7-------4--------3--------0--------7
Militia Archers--------------2/5------0--------1--------0--------1
The Swordsmen, Knights and Longbowmen were fully upgraded. The rest had no upgrades.

All averages reported below (unless otherwise stated) are from a minimum of 15 tests for each scenario.

Attacking Unit Type
Three spies were sent into the Castle to give 103% chance of opening the gates. The game was quick saved / quick loaded and the settlement attacked.

Attack on Castle using 601 men
Attacking Unit Type-----------------Lost----------Kills
Armoured Swordsmen-------------87---------- 240
Feudal Knights-------------------107----------232
Retinue Longbowmen-------------141----------234

Town Militia----------------------311----------52------defeat
Hobilar---------------------------336----------95------defeat
Militia Archers--------------------411----------66------defeat

The three higher quality units all had similar kill and capture rates but the losses showed significant differences. Infantry < Cavalry < Missile. How these figures are calculated is a mystery. It is tempting to argue that the kill rates of the heavier units are reflecting their attack value. But this does not hold true for the lighter units. Similarly the defence values do not follow the same pattern as the losses.
All the attacks with the heavier units gave victories whereas all the lighter units gave defeats. When I repeated these attacks with full stacks of light unit (1500/800/1200) they also gave defeats, but with lower losses.
Smaller sample size tests with mixed armies (i.e. 750 Town Militia + 600 Militia Archers) showed no advantage.

Generals
Three spies were sent into the Castle to give 103% chance of opening the gates. The game was quick saved / quick loaded and the settlement attacked. One of two generals was added to the army.

King Robin Faction Leader
Command = 12, Dread = 6, Authority = 11, piety = 4, HP = +9, Unit size 38

Richard Blake
Command = 0, Chivalry = 1,, piety = 0, HP = +0, Unit size 20

Attack on Castle using 361 Armoured Swordsmen.
General--------------Lost---------------Kills
None--------------120---------------241
Richard Blake------101---------------242
King Robin---------69----------------255
King Robin---------63----------------252

The addition of even a low ranking general gives a significant improvement to the losses that cannot be readily attributed to the 20 extra men of his body guard. The higher ranking general reduces the losses to half that expected with no general. The Night Fighter trait, of the faction leader, gives a +1 to command at night but since he is already maxed out in command this might not be taken into account in the calculations. The reduction in night battles losses could be due to a penalty against the accuracy of the defending missile units.

Castle Vs Town
Three spies were sent into the settlement to give 103% chance of opening the gates. The game was quick saved / quick loaded and the settlement attacked.

Attack on Castle and Town using 361 Armoured Swordsmen
Attack---------------Lost------Kills
Castle-------------120------241
Town--------------94-------235

Expect 20 - 25% less casualties when attacking towns.


SIEGE ENGINE TYPE
Three spies were sent into the settlement to give 103% chance of opening the gates. The game was quick saved / quick loaded and the settlement attacked. Then the game was reloaded the spies removed and the siege started without selecting any siege equipment. The game was quick saved and quick load, the siege equipment selected and the turn ended. The settlement was then attacked at the start of the next turn.

Attack on Town using 361 Armoured Swordsmen
Method--------------Lost------Kills
3 Spies------------94-------235
1 Ladder-----------102------217
1 Ram-------------187------237

Increasing the number of spies improves the chance of opening the gates but does not affect the losses and kills. Once the gates are open there is the same results with 1 spy as with 9 spies.

Although I have only reported above the results with armoured swordsmen I have done testing with a variety of unit types and number of siege engines. The above shows clearly that the losses using Spies < Ladders <Rams. This result is also repeated when sieging castles and also when using Retinue Longbowmen and different size attacking forces.

The number of siege engines that you build has no effect on the result. You get exactly the same set of results if you build 1, 2, 3, 4, 5, or 6 set of Ladders. This means that not only does the calculation treat any number of Ladders in the same way but it does not even affect the random numbers used to give different results.
The same is true for Siege Towers. Not only does the calculation treat the number of Siege Towers in the same way, it also treats them the same as Ladders.
You will get exactly the same set of results if you build
1 Ladder.
3 Ladders
1 Siege Tower.
2 Siege Towers and 3 ladders
Or any combination of Towers and Ladders.

Rams are similar in that the number you build always gives the same result but the result is different from that of Ladder/Towers and always gives higher losses. If you build a combination of Rams and Ladder/Towers the calculation ignores the presence of the Rams and returns the result expected from building a Ladder.

I only reported the results for one Ladder and one Ram, because all the other results are exactly the same.

Summery
Most of the work above just confirms what most experience players have already worked out for themselves. That is -
Spies are the best “siege engine”.
If no spies are available, then just build one set of ladders. Anything more is wasted.
Have the best general available to lead your attack (preferably at night).
The higher the quality of the units, the better the results.
Choose infantry over cavalry over missile units.
Expect a tougher fight for a castle than for a town.
Think twice about whether you need to auto resolve. You may obtain better figures and a quicker result. But you won’t gain experience and you will miss out on a lot of fun.

Finally
There is one major factor to auto resolving sieges (or anything in this game) that I have hardly touched on. The random generator!
Almost any calculation or decision you ask the game to make involves two elements. The first is a normal mathematical equation that can precisely calculate the losses. The second is the random factor that CA included in all the calculations to make the game less predictable and more fun. Unfortunately this random factor also makes the sort of comparisons reported above a lot harder. Whilst in a few cases it is possible to control the game so that the same random numbers are used in two scenarios that are being compared, for most this cannot be done. For these each part of the test has to be repeated until there is a big enough sample to average out this random element.
Whist doing this you realise just how big a difference these random factors can make to the results. For example whilst calculating an attack on a Castle using 361 Armoured Swordsmen, the sample size was 45 tests. This gave the following range of results -

---------------------Lost------Kills
Maximum----------180------155
Average-----------120------241
Minimum-----------80-------294

If the results can vary so much, due to the random number when carrying out exactly the same siege on the same castle in the same turn etc. Then just how much time and effort is it worth spending selecting the best units and tactics for the job? How much is the success in a campaign due to your skill as a commander and how much to the roll of a dice?

[Sentinel]

ARRRR !
Could some nice kind moderator type person please correct my title for me? It should read.
Auto Resolving Sieges

Thanks

[Foz]

Since the settlements have been garrisoned with archer units, it would make sense to include in your analysis the applicable defense against missile fire that each unit involved has. You note that the results do not directly follow from each unit's attack values, nor their defense values, but perhaps a combination of attack and missile defense DOES give a better sense of how the unit will perform against a settlement full of archers. You noted that armoured swordsmen, then feudal knights, then Retinue Longbowmen was the order of preference for heavier units. Their missile defense stats look like this respectively: 14, 11, and 8. Looks to me like it's directly tied to how many losses the unit takes. The lower tier units, Town Militia, Hobilars, and Militia Archers, have missile defense 6, 4, and 0... so it seems very clear that all of your results are due to how resistant the given unit is to the missile fire the enemy units in the garrison are laying down.

[sapi]

Interesting.

@foz - that could be a factor, yes

@sentinel - i've fixed the title for you

[Sentinel]

Sapi - Thanks for fixing the title.

Foz- What you say makes sense.

To confirm the theory I ran the test again with Bill Militia, that also has a total missile resistance of 0. The result were

Attacking Unit type-----------Lost----------Kills
Bill Militia--------------------391----------156

This fits the theory.

This made me wonder what results melee defenders would give. So I retested this time giving Magdeburg to the Papal States who garrisoned it with -

4 units of Mailed Knight.
2 units of Mounted Sergeants
2 units of Dismounted Feudal Knights.
All Melee - No Ranged units

The results were -

Attacking Unit Type-----------------Lost----------Kills

Armoured Swordsmen--------------59--------- 202
Feudal Knights--------------------88----------213
Retinue Longbowmen-------------129----------208

Town Militia----------------------288----------100------defeat
Hobilar---------------------------317----------84------defeat
Militia Archers--------------------354----------50------defeat

This is the same order as for the ranged defenders. Maybe because the settlement has so many towers, that inflict missile damage, the calculation assumes that all the defenders inflict missile damage?

[Foz]

Hmm. As always I'm back with the defense analysis, this time for melee.

Armored Swordsmen: 22
Feudal Knights: 16
Retinue Longbowmen: 14
Town Militia: 7
Hobilars: 7
Peasant Archers: 1

They fall in exactly the same order as the missile defenses of the units did, I believe. So you can't say that the game is calling the damage from the melee defenders missile damage, there's simply no evidence that points to that. Granted I just read all those numbers out of the file, and notice that several do not agree with what you quoted in your original post. How exactly is it that your Feudal Knights and Retinue Longbowmen came out with the stats you listed? (btw you added the feudal knights def up incorrectly, 8+5+5=18).

It'd also point out that though town militia and hobilars have the same melee defense, there are several reasons the town militia probably beat hobilars:

1. They have many more men in a unit, and on the field, at one time. This allows them to have more men poking at the enemy unit than hobilars could, and thus possibly killing a few more.

2. They have anti-cavalry properties, and as most of the melee units garrisoned are cav, this may play some part. Also if routing and morale are considerations, they will rout with more men left due to having a larger unit size than hobilars, and thus more men are likely to successfully scurry off the battlefield (and so not be killed or captured). This may not happen so much at the end of the battle when cavalry can give chase, but in the beginning one would expect a fresh batch of militia enters the fray to replace the routing one, and they largely escape.

One more thing to point out - note how, while taking nearly the same number of casualties, the bill militia cause considerably more kills than do any other low-tier unit - almost up to the level of the upper tier ones. Just a guess here, but it's probably due to their much better attack value. They don't kill as many as upper tier units with comparable attack values, though, which seems to indicate that the attack and defense are somehow blended to determine the effectiveness of the unit overall. So I'd suggest that kills and losses aren't generated separately, but rather are results of some calculations that determine the direct unit strengths versus each other.

[KHPike]

Don't know if it's been mentioned, but I think auto-resolve causes the game to ignore settlement defenses. It also ignores advantages conferred onto defenders when on the walls I think, and this minimises your casualties. I don't have the exact stats here now, but I tried comparing auto-resolve vs assaulting myself.

I assaulted Tolouse as Milan with 5 Genoese Archers, 4 Famiglia Ducale, a general and the rest dismounted broken lances (full stack) vs a French army with lots of sergeant spears, some knights and archers (castle with full stack)

When I auto-resolved I lost only about 400 men. However when I played it out myself it was a near disaster as I lost 50% of my men to the walls (maybe I'm a bad assaulter?) The Frenchies burnt down one siege tower and my ram...

[Nebuchadnezzar]

You didn't mention if you had any fixes eg 2H fix. This could make a world of difference in favour of playing out seiges.

I auto resolve more than 70% of my battles to avoid the passive AI and also because they get very repetitive, particularly sieges. I find that I would usually sustain only moderately more casualties fighting a seige than auto-resolve, but not always. On the other hand sometimes it can be a total disaster if you loose your seige equipment.

Reloading and replaying an auto-seige will give you a very different and better result again!

[Sentinel]

KHPike
I found differences in the losses, when attacking a castle and town. Therefore the calculation cannot be completely ignoring the settlements defences. But you may be right that it does not place sufficient emphasis on the defences.

Given the odds you faced in your Toulouse assault I think you did quite well. I doubt if I could have done any better.

Nebuchadnezzar
I run the original game with patch only on VH/VH. No other mods, fixes or cheats. I am waiting until after the next patch before considering if any fixes are required.

I also have used auto resolve more in this game than all the other games put together. In M2TW you often achieve more favourable results, especially in the early game or when using a larger attacking force of low quality troops against a small garrison. This is partly due to the new calculations used and partly due to the losses from the tower defences when manual fighting the game. In the later games, auto resolving has less an advantage, due to a bigger choice of siege equipment available and therefore greater flexibility to match your tactics to counter the defences.

Foz
Unit Stats.
I have rechecked my original post and you are correct there is an error.
The Town Militia, Hobilars, and Militia Achers are all standard units with no upgrades. The Armoured Swordsmen have weapon and armour upgrades giving them +1 to attack and +1 to armour defence. The Retinue Longbowmen has heavy mail armour and also experience 2 (from the woodmen guild HQ).
All the above details were correctly quoted.

The Feudal Knights also had weapon and armour upgrades, which gave them
Attack 11, armour 8, skill 5, shield 4, Total 17
And not
Attack 11, armour 8, skill 5, shield 5, Total 17 as I quoted. Sorry!

Hobilars vs Town Militia

Unit size
This could be argued both ways.. The town militia has more men per unit (8 x 75) but the Hobilars have more units (15 x 40). Both attacks used 600 men.

Against melee garrison
The Town Militia has a spear bonus + 4 verses cavalry (2/3rds of garrison) making attack 9. This does not apply to infantry (1/3rd garrison), leaving them 5 with these units. This would average out at 7.67 giving them a 0.67 advantage over the Hobilars..
Their total defence is the same as the Hobilars but mainly in shield, which only applies to front/left. If attacked from the other two sides they have almost no protection. Hobilar defence is mainly the more effective armour (applies 360°).

Against missile garrison
The Hobilars hits with 7 giving them a 2 advantage over Town Militia. Since the crossbow missiles are armour piercing, this gives an effective protection of 2 (360°) for the Hobilars verses 3 (front/left) for the Town Militia.

Both have a base moral of 3. Both will rout when they have lost a certain percentage of their numbers. Since the Hobilars are faster, they are less likely to suffer further losses whilst running away. Their speed will also get them past the towers / missile fire into a melee with fewer losses.

There is not a lot of difference between these units. Personally I would have put my money on the speed and flexibility Hobilars for both attacks. Whichever way you calculate it the Town Militia should not be better than Hobilars for both attacks.
The only theory that does fit the order of losses is to assume that the defenders are missile and the losses are ordered according to their missile resistance. This may not be an ideal explanation, but so far it is the only one that fits.

One of the problems with tying to find out what factors CA use in these calculations is the random numbers applied to the results. This suggests that there is not much point in CA spending a lot of time and effort (therefore money) developing complex formulae. Why calculate a units losses to the second decimal place, and then apply a massive random factor that make all the accuracy worthless.
They are much more likely to do a rough defence / attack average for both sides, season it with a few unit size / environment / unit type factors. Stir it all up with a set of random numbers and serve with a fanfare. Whilst the player simmers at the results.
Grouping all defenders (with towers) into one type of attack (missile) would simplify the calculation. Would they really go to the trouble of differentiating and allotting different factors to all the various types of attack.
e.g. Melee, spear bonus 4, spear bonus 8, armour piercing, normal arrows, ballista bolt, exploding shot, phalanx etc. etc. etc.