Game Mechanics: AC may not be working correctly

[Splorf22]

I did this test for the SK shield thread. The error bars aren't small enough to be really sure (and I'm not going to do the 3 hour test that would require) but the basic point is that equipping a 40AC shield was somewhere between worthless and actively bad for mitigation even over a decently sized test. 40 item ac (70 displayed AC) is a huge, huge amount: it's basically the equivalent to wearing a Kunark tank BP and going commando.

Originally Posted by Loraen

I did a test with Sakuragi vs the Cliff Golem in OT with and without the Shield of Elders (40 item AC). Both tests were about 5 minutes and had about 150 hits with various damage values. I then bootstrapped the data (my new favorite technique) to get:

1120: mean: 113.5 +- 3.4 [95% CI: 106.6 120.4]
1050: mean: 111.3 +- 3.9 [95% CI: 103.6 119.1]
1120 > 1050: 66.7%

What this means is that with 1120AC, I expect the average hit to be around 113.5 and almost certainly in the range 106-120. With 1050AC, I expect the average hit to be 111.3 (yes, this is LESS. I double checked to make sure I didn't mix up the parses) and almost certainly in the range 104-119. The probability that you will take more damage with 1120 AC is 67%.

The variance in this data is quite high. From what I understand, hits are either min, max, or a purely random hit in between. So I modified the data appropriately, which removes some of the randomness:

Replacing any number not 48 or 172 by 109.5
1120: mean: 111.8 +- 2.6 [95% CI: 106.6 117.1]
1050: mean: 111.5 +- 3.0 [95% CI: 105.6 117.5]
1120 > 1050: 52.0%

Thus we can see that the 1050 parse was a bit luckier on the damage interval hits, but the variance is still quite large. The variance goes down as the square root of the number of samples, so we'd need about 50 times as much data (3 hours with each configuration) to get a confidence interval with a size of +- 1. Needless to say, I don't think I can rent Coeur for 6 hours.

My conclusions:
1. Someone not named Sakuragi should tank the cliff golem for a few hours
2. Pasi is correct; for raids HP >> AC. Not only is the average value of AC quite small, but you care about about the worst case more. Based on this test if you are a main tank I would not give up even one HP for additional AC. This is actually good news for Iksar warriors!
3. Even if AC is working correctly (which is only a 50% chance according to this data, heh) the benefit of a ridiculous +40AC shield is very unlikely to be more than a few percent for high-level mobs.

Of course, all of this may be different for xp mobs with lower attack values. For reference, without the shield Sakuragi has about 170 raw item AC.