Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Drazinononda]

The difference between heavy and light breastplates is the resultant acceleration from a projectile strike. Assuming real-life physics for the sake of the thought experiment: a silver bolt striking a steel breastplate would cause a small acceleration, in relation to the high mass of the breastplate, because a=F/m; the same bolt striking an adamantine breastplate would cause a large acceleration conversely.

Keep in mind that almost all effects of damage that we know of are related more to acceleration than to mass, velocity or material strength. For example, a person may be able to lift a 100kg barbell without any ill effect, but if you smacked them across the palm sharply with a wooden dowel you could bruise the skin or even conceivably fracture a bone. The difference is not necessarily the amount of force applied, but the amount of time over which it is applied: which is why it is perfectly safe to take off in an airplane and reach 400km/h over the course of several minutes, but not to be fired from a cannon and reach that speed in less than a second. Accelerations higher than the tensile strength of a tissue will cause damage, whether that acceleration is due to a blunt object or gravity or whatever.

Keep in mind also that this is more important at the boundary between the armor and the body inside it than between the armor and the projectile, in cases where the projectile doesn't penetrate the armor at all. If you were only testing bolts vs. armor, steel and adamantine would be equally effective against any bolt softer than either of them, but the difference on the skinward side comes out more as the difference between a hard shove and a heavy punch.

Of course, all bets are off if bolts can in fact "punch holes" in adamantine shaped armor... I don't mess with the bluemetal enough to comment there.

[Urist Da Vinci]

...
What about the law of conservation of energy? ...

http://dwarffortresswiki.org/index.php/DF2012:Water_wheel#Perpetual_motion

I also like how the velocity of thrown objects is a constant (100, compared to the default 1000 of crossbows), so you can throw a giant sperm whale corpse at someone (with no recoil) and the momentum is enough to send the target flying backwards.

So to optimize armor layering, you want to know how the game actually treats armor, not how it should work if it followed reality.

We only tested bolts. Melee weapons probably follow similar logic, but there are extra combat variables to consider, so melee combat can be more complex.

A good scientist never assumes, they test and test until they find the actual answer. It's fine to have hypothesis but taking it as fact isn't good.
...

Look at how complex the 40d damage formula was, and that was when materials were simplified: http://dwarffortresswiki.org/index.php/40d:Weapon#Damage_Formula
I can't think of a good experiment design for melee combat. Perhaps a modified version of the mummy curse can be used to eliminate some of the random dice rolls.

[Centigrade]

There sure are a lot of words, none of them directly saying the optimal armour layout.

[vanatteveldt]

Hmm, I just had a look at the data presented in the excellent 'The end of "What bolt metal is better?" arguments' thread linked in Urist's sig.

What I take from it is not that candy is worse than steel. It performs similar against everything except wood bolts, which are not your biggest enemy in any case, and wood bolts are still really bad against candy. Against candy or bone bolts candy actually outperforms steel, presumably because these bolts have so little mass that they depend on penetrating the armor.

Is there any reason to assume that the "wood bolt syndrome" translates to melee cases? Has anybody done an arena experiment to pit 1000 candy/steel/mix clad dwarfs against a variety of actual opponents (say goblin, minotaur, clown or something) and compare survival rates? Of course there is a lot of random but with only 9 cells and >100 cases per cell it shouldn't be too bad, I suppose? You can also have an unarmed dwarf and count number of blows before injury / incapacitation / death?

(I've never used the arena and have never seen candy other than by throwing my miner into a volcano (it was an accident, I swear!), so I don't think I am the person for doing this... I'll happily parse the logfile if someone can do the arena part, I'm good at scripts and re?)

[Button]

There sure are a lot of words, none of them directly saying the optimal armour layout.

Why, it's almost as if no one knows for sure, there's disagreement on the issue, and they're trying to work it out.

[DWC]

Ack, I really wish the game still used the 40d combat with maybe some minor tweaks. This material science model is too ambitious maybe, too much of it does not reflect reality better then the old hit-point system did.

[Loud Whispers]

There sure are a lot of words, none of them directly saying the optimal armour layout.

Why, it's almost as if no one knows for sure, there's disagreement on the issue, and they're trying to work it out.

Science: We're all wrong, but we'll always try to be less wrong than yesterday.

[Centigrade]

There sure are a lot of words, none of them directly saying the optimal armour layout.

Why, it's almost as if no one knows for sure, there's disagreement on the issue, and they're trying to work it out.

I mean that no one is even putting forth hypotheses; instead, they're talking about materials. Here, let me show you.

Let us define dwarfs wearing Adamantine breastplates as population 1 and dwarfs wearing Steel breastplates as population 2. Let us assume that both populations of dwarfs are otherwise identically equipped and skilled (specify further), and let us design an experiment to discover if Adamantine breastplates offer greater survivability than Steel breastplates. We shall elect to determine survivability by examining the average number of melee strikes (specify further) required to slay a dwarf wearing each type of breastplate. By taking several independent samples and comparing the gathered data, we may test the hypothesis that Adamantine breastplates offer greater survivability than Steel breastplates.

Hypothesis₀: μ₁ = μ₂
Hypothesis_A: μ₁ > μ₂

[vanatteveldt]

Well, without the fancy greek and subscript notation, I think I suggested an experiment with exactly that (implicit) hypothesis.

I think lack of science is the problem, not lack of words...

[Centigrade]

Well, without the fancy greek and subscript notation, I think I suggested an experiment with exactly that (implicit) hypothesis.

I think lack of science is the problem, not lack of words...

I am willing to help with the testing. I have a pretty great machine, as well as a spare one that does literally nothing all day, that I could use for collecting data. I even have a background in statistics, so I would be happy to help analyse the data. I do not have a background in graphic design, so my analysis would consist of walls of numbers and/or text rather than pretty graphs like the ones used in the Bolts v. Armour research project. But before we get started on the research, we need to first define an exhaustive series of tests to be conducted, as well as control variables for these tests. For lack of a better metric, I am going to put forth an arbitrary armour set which I will use as the control for future testing, which I will generally designate as population 1.

Steel and Pigtail Fiber (Control Population)

Spoiler (click to show/hide)

☼steel helm☼
☼pigtail fiber hood☼
☼pigtail fiber hood☼
☼pigtail fiber hood☼
☼pigtail fiber hood☼
☼pigtail fiber hood☼
☼pigtail fiber hood☼
☼steel mail shirt☼
☼steel mail shirt☼
☼steel mail shirt☼
☼steel breastplate☼
☼pigtail fiber cloak☼
☼pigtail fiber cloak☼
☼pigtail fiber cloak☼
☼pigtail fiber cloak☼
☼pigtail fiber cloak☼
☼pigtail fiber cloak☼
☼steel gauntlets☼
☼pigtail fiber mittens☼
☼steel greaves☼
☼pigtail fiber trousers☼
☼pigtail fiber trousers☼
☼steel high boots☼
☼pigtail fiber socks☼

I do not include a shield or weapons in the armour set, because blocks and parries add unnecessary variables to the testing of the rigid armour. I will similarly use dwarfs with no skill in armour use or dodger, so that the variable being tested shall be isolated. In an actual combat situation, a given dwarf would obviously want as high skill as possible in armour user and dodger, and to have a shield and one or more weapons, to provide extra opportunities to reduce incoming damage.

Our first step would be to establish a large volume of data about this control population. Ideally multiple samples with large sample sizes would be run through the arena, against groups of enemies in varying numbers and of varying types. This is the part of the testing that makes me somewhat wary, because I do not know how best to simulate a combat encounter. While a simple 100 v. 100 death match in which we record how long it takes for the dwarfs to eventually die to 100 stray kittens (or demons, or whatever) might give us a large volume of data, I wonder at the specificity of the data. I also wonder which enemies we should test. While we could conceivably test every single enemy, for my initial testing I am going to limit the attackers to humanoids. I am thinking that perhaps we should for each run of the experiment, that is for each armour set we want to test, we should perhaps do individual tests against battle axes, great axes, flails, halberds, large daggers, long swords, maces, mauls, picks, pikes, scimitars, scourges, short swords, spears, two handed swords, war hammers, and whips; and, that each of these tests should be conducted twice, first with the attackers at the Proficient skill and second with the attacks at Great skill. This would give us 34 data sets, for a given run of the experiment.

This is all just to establish a single sampling of the control population, for the above Steel and Plant Fiber armour set. Ideally the exact same experiment would be run several times so that we would have multiple samples from which to extrapolate population statistics. Conducing actual testing would require that an alternate armour set, preferably changing only one variable at a time (i.e., plant fiber clothes v. silk clothes, adamantine rigid armour v. steel rigid armour, or adamantine mail shirts v. steel mail shirts) be created and run through the same experiment; again, more samples is better so that we can extrapolate population statistics. At the end of several rounds of testing, once we have extrapolated population statistics rather than sample statistics, we would arrive at a falsifiable hypothesis; for example, "the mean number of short sword strikes required to strike down a dwarf wearing steel and plant fiber at the Proficient skill level (μ_1A=3) is greater than the mean number of short sword strikes required to strike down a dwarf wearing steel and plant fiber at the Great skill level (μ_1B=2)." That much is intuitive, and ought to be true; i.e., an attacker who is better at hitting things requires less hits to kill a thing. We would make further comparisons against every population statistic we calculate.

It also just occurred to me that to be really thorough, we ought to replicate the test several times: a full run each for a few different attacking weapon materials, e.g., copper, iron, steel, and adamantine. That would give us 136 individual 100 v. 100 simulations for a single sample.

edit: All this testing could take a while, since as far as I know you have to individually add creatures one at a time.

[Centigrade]

With regards to the above post, I have set out to begin testing on the control population, which will require me to individually place 200 creatures 136 times for the first sample. That's a lot of button pressing that will probably take me the better part of a couple of days. I would greatly appreciate if anyone else interested in this topic would be willing to run a sample themselves, so that we can get more data collected in a short amount of time. Just to be clear and so that we are all using the same methods, I am spawning individual creatures in two teams, in the central, circular-shaped room surrounded by water on the default object testing arena; team 1 on the left, team 2 on the right. I am using the unmodified DF2012 v0.34.11 game client for these tests.

Spoiler (click to show/hide)

I am using humans as the attackers in all tests, because they are large enough to use all the weapons needed. Humans are assigned to team 1, on the left side. I am using dwarfs as the defenders in all tests, because they are the default race for fortress mode. Dwarfs are assigned to team 2, on the right side. I am individually spawning 100 of each creature on its appropriate side and team, in a single tile, then unpausing the game and exporting the gamelog.txt file for analysis. I will of course provide that file as raw data for others to review. Not that I believe it matters, but I am using exclusively female combatants in this test, because that's one less keystroke per creature.

Spoiler (click to show/hide)

edit: Images go in spoiler tags.

[Loud Whispers]

Multiple series of 1vs1 duels would yield more reliable results than 100vs100.

[Centigrade]

How so? We're dealing with how many hits it takes to kill an unarmed but armoured dwarf. Why would it matter whether he takes five hits from one human or one hit each from five humans?

[Centigrade]

My first few runs of the test, with the attackers wielding copper weapons, have been done, and I am noticing something that bugs me. Frequently the test will end with one or more dwarfs alive but exhausted/unconscious and apparently invincible because they are wearing helmets and the enemies are exclusively targeting the head. How should I factor this into my analysis later?

[Loud Whispers]

How so? We're dealing with how many hits it takes to kill an unarmed but armoured dwarf. Why would it matter whether he takes five hits from one human or one hit each from five humans?

Snowball effects can really mess up these sort of tests, especially on a scale of hundreds.