Treasure Hunter Testing

The problem with any kind of multiple items dropping is that we don't know how multiple items work at all. They could be independant, they could be not. It could work tiered too, which is what I consider most likely, similar to (if not the same as) armoury crates. For each slot there's several options that can drop. Whether or not a slot drops at all could be affected by TH, the rest distributed (not necessarily uniformly) within that slot. Or, there's several groupings, some of which allow several drops, and the number of drops is affected by TH, possibly even whether or not that grouping is considered for drops at all.

In short, any mob with a too complicated drop pool should be used as a testing subject, because we don't know how to interpret the results. An ideal starting testing mob would be one with only one drop. The Colibri's drops seem largely independent (although it's impossible to confirm), since the very large samples indicate that they both behave the same way under the influence of TH. Once we have TH figured out for very simple groupings (there's not many logical combinations with only 2 drops), we can approach more complex structures and analyze how they behave compared to the existing data.

Do you have more accurate testing data for those Dhalmel drop rates, as in, item groupings and their distribution? That could be very relevant, otherwise someone will have to repeat that test again. Can't even say if Dhalmel Hair drops are unrelated to the others.

Anyone wanna go mass murder Rock Lizards in South Gustaberg? A quick glance at wiki shows they have only one drop, with kinda low drop rate, easy kills and tons of them all around. Can even help noobs get some boots while at it. I think that's what I'll be doing once I get to play again. Would be nice to see if they confirm the Colibri findings, then we can use those as a base for mobs with two independant drops. From there on, we could try to confirm other multi-drop mobs' treasure pools and how they're affected by TH.

Sorry, in a rush at the moment, but did you account for the fact that multiple different items can drop from the same drop slot? The problem with using a monster as messy as Dhalmels is that you don't necessarily know all drops are independent.

Edit: To explain a little more...
Slot 1: Dhalmel Meat or Dhalmel Hide
Slot 2: Dhalmel Meat or Papaka Grass
Slot 3: Dhalmel Hair or Giant Femur

Assuming TH "re-rolls a slot," the above situation would give dramatically different result than:
Slot 1: Dhalmel Meat
Slot 2: Dhalmel Meat
Slot 3: Dhalmel Hide
Slot 4: Papaka Grass
Slot 5: Dhalmel Hair
Slot 6: Giant Femur

iirc though, in the past when I've run the numbers on various situations, the increase is always higher than I'd predict with multiple drops sharing the same slot. At the end of the last thread on this, I just threw up my hands and said, "This looks like maybe how it works, but not for things that have multiple drops per slot."

Quote:

---

Originally Posted by Babekeke

All I can say to this is OMFG, are you serious?
If they reported back and said: we'd have sent them back with their tails between their legs. Instead, they produce a fair sample size, and idiots like you just see it as NEEDING 1000 kills to see a difference.

The trouble with increasing a chance, is that it's still a chance. If you increase TH to have a 95% drop rate and kill a mob once, you could still not get the drop then next time kill it with 5% chance of drop and it drops.

I hope you're glad I took out the time to respond to your trolling :)

---

The test required a sample size that big. You are a moron.

Edit: to reiterate. They didn't have to kill 1500 mobs to figure out how much of an increase it was. They killed 1500 mobs just to have reliable proof that there was any increase in drops at all.

Myth about MNK gets TH +9 and +10 gear SWEET.... I seen it in a test image, and it's already fully developed it must be true!

You know this is how things get started right!

Here's the grouping data for those parses:

Code:

---

TH0:
      Dropped  0 items  785 times ( 78.34 %)
      Dropped  1 items  189 times ( 18.86 %)
      Dropped  2 items    23 times (  2.30 %)
      Dropped  3 items    5 times (  0.50 %)


    Number of times each group of items dropped.

      785 [ 78.34 %] -- Nothing
      101 [ 10.08 %] -- tuft of dhalmel hair
      68 [  6.79 %] -- slice of dhalmel meat
      12 [  1.20 %] -- giant femur
        8 [  0.80 %] -- slice of dhalmel meat, slice of dhalmel meat
        7 [  0.70 %] -- dhalmel hide
        6 [  0.60 %] -- slice of dhalmel meat, tuft of dhalmel hair
        4 [  0.40 %] -- giant femur, slice of dhalmel meat
        3 [  0.30 %] -- giant femur, tuft of dhalmel hair
        2 [  0.20 %] -- giant femur, slice of dhalmel meat, tuft of dhalmel hair
        1 [  0.10 %] -- dhalmel hide, sprig of papaka grass
        1 [  0.10 %] -- sprig of papaka grass
        1 [  0.10 %] -- dhalmel hide, dhalmel hide
        1 [  0.10 %] -- slice of dhalmel meat, slice of dhalmel meat, tuft of dhalmel hair
        1 [  0.10 %] -- dhalmel hide, sprig of papaka grass, tuft of dhalmel hair
        1 [  0.10 %] -- giant femur, slice of dhalmel meat, slice of dhalmel meat

TH1:
      Dropped  0 items  598 times ( 59.80 %)
      Dropped  1 items  272 times ( 27.20 %)
      Dropped  2 items  101 times ( 10.10 %)
      Dropped  3 items    29 times (  2.90 %)


    Number of times each group of items dropped.

      598 [ 59.80 %] -- Nothing
      143 [ 14.30 %] -- slice of dhalmel meat
      101 [ 10.10 %] -- tuft of dhalmel hair
      57 [  5.70 %] -- slice of dhalmel meat, slice of dhalmel meat
      22 [  2.20 %] -- giant femur, slice of dhalmel meat
      18 [  1.80 %] -- slice of dhalmel meat, tuft of dhalmel hair
      17 [  1.70 %] -- slice of dhalmel meat, slice of dhalmel meat, tuft of dhalmel hair
      12 [  1.20 %] -- dhalmel hide
      12 [  1.20 %] -- giant femur
        6 [  0.60 %] -- giant femur, slice of dhalmel meat, slice of dhalmel meat
        4 [  0.40 %] -- sprig of papaka grass
        3 [  0.30 %] -- giant femur, slice of dhalmel meat, tuft of dhalmel hair
        2 [  0.20 %] -- dhalmel hide, slice of dhalmel meat
        1 [  0.10 %] -- dhalmel hide, sprig of papaka grass
        1 [  0.10 %] -- dhalmel hide, giant femur, slice of dhalmel meat
        1 [  0.10 %] -- giant femur, slice of dhalmel meat, sprig of papaka grass
        1 [  0.10 %] -- dhalmel hide, dhalmel hide, dhalmel hide
        1 [  0.10 %] -- giant femur, tuft of dhalmel hair

TH2:
      Dropped  0 items    97 times ( 48.50 %)
      Dropped  1 items    66 times ( 33.00 %)
      Dropped  2 items    26 times ( 13.00 %)
      Dropped  3 items    7 times (  3.50 %)
      Dropped  4 items    4 times (  2.00 %)

    Number of times each group of items dropped.

      97 [ 48.50 %] -- Nothing
      41 [ 20.50 %] -- slice of dhalmel meat
      21 [ 10.50 %] -- tuft of dhalmel hair
      14 [  7.00 %] -- slice of dhalmel meat, slice of dhalmel meat
        6 [  3.00 %] -- slice of dhalmel meat, tuft of dhalmel hair
        4 [  2.00 %] -- giant femur, slice of dhalmel meat
        4 [  2.00 %] -- giant femur, slice of dhalmel meat, slice of dhalmel meat, tuft of dhalmel hair
        3 [  1.50 %] -- slice of dhalmel meat, slice of dhalmel meat, tuft of dhalmel hair
        2 [  1.00 %] -- giant femur, tuft of dhalmel hair
        2 [  1.00 %] -- dhalmel hide
        2 [  1.00 %] -- giant femur
        1 [  0.50 %] -- dhalmel hide, slice of dhalmel meat, tuft of dhalmel hair
        1 [  0.50 %] -- dhalmel hide, dhalmel hide, sprig of papaka grass
        1 [  0.50 %] -- giant femur, slice of dhalmel meat, tuft of dhalmel hair
        1 [  0.50 %] -- dhalmel hide, dhalmel hide, slice of dhalmel meat

---

I got a maximum of 4 drops with TH2. Given the probabilities of the highest dropping items (24% for each meat, 19% for hair, 7% for femur), the best chance of getting a 5-item drop pool assuming all drops are fully independant is 0.018%, or 1 in 5500. Since I only killed 200 in TH2, and 1000 in the others, it is not at all surprising that I didn't see that many at once.

The odds of a femur/meat/meat/hair drop (the only 4-drop combo) given the above probabilities is 0.17%. I 'should' have seen such a combo only once in 570 kills, but instead got 4 in 200. 0.17% is actually outside the confidence interval for 4/200, so there's certainly more to be considered.


Given possible drops of:
dhalmel meat (x2)
dhalmel hide (x3)
dhalmel hair
giant femur
papaka grass

A total of 8 items, which should be perfectly fine within the known limits of the game.

Combinations of items include:
femur/meat/meat/hair -- femur, meats and hair are all independant
hide/hide/grass -- grass is independant of at least 2 of the hide drops
hide/grass/hair -- hair is independant of at least one of the hide drops as well as grass
hide/femur/meat -- one hide is independant of femur and at least one of the meat drops
femur/meat/grass -- grass is independant of femur and at least one of the meats
hide/hide/meat -- at least one of the meats is independant of two hides


Even assuming some of the items can share drop slots, at best you're getting the third hide shared with the femur, the grass, the hair or the second meat. Such a limited combination seems rather pointless. I could maybe see the third hide being shared with the hair since the hair is a new drop for those mobs compared to previous dhalmels, but since the hides aren't part of the main analysis comparison here due to their relatively low drop rate it wouldn't matter much.


Reviewing the meat distributions:

TH0 data:
100 total
90 instance (kills with at least 1 meat)
10 2x drops

Raw drop rate: 100/1002 = 10%
Drop rate of total/(kills*slots) = 100/2004 = 5.0% (4.1%-6.0%)
Unique drop rate = 90/1002 = 9.0% (7.4%-10.9%)
Ratio of double to instance: 10/90 = 11.1% (6%-19.4%)


TH1 data:
350 total
270 instance
80 2x drops

Raw drop rate: 350/1000 = 35%
Drop rate of total/(kills*slots) = 350/2000 = 17.5% (15.9%-19.2%)
Unique drop rate = 270/1000 = 27.0% (24.3%-29.8%)
Ratio of double to instance: 80/270 = 29.7% (24.5%-35.3%)


TH2 data:
96 total
75 instance
21 2x drop

Raw drop rate: 96/200 = 48%
Drop rate of total/(kills*slots) = 96/400 = 24% (20%-28%)
Unique drop rate = 75/200 = 37.5% (31.1%-41.4%)
Ratio of double to instance: 21/75 = 28% (19%-39%)



You'll note that in each instance, the rate of double drops is higher than it should be if they were fully independent (compare double-to-instance ratio with total-per-kill*shot ratio). Margin of error for TH0 and TH1 also don't overlap between the two measures. TH2 overlaps, but 2x drop ratio is mostly higher than nominal drop rate.


Suppose we assume that the second meat can only drop if the first meat drops. That is, they are not two independent slots, but one independent slot and one dependent slot.

The drop rate on the independent slot can be determined by the number of unique kills where that item appeared (unique-drop-rate ratio). In this case, that gives 9%, 27% and 37.5% for TH0/1/2, respectively.

In that case, the odds of the second drop appear to always be within the margin of error of the unique drop rate ratio.

Implication: this is where multiple rolls come in. EG: Chance of main drop is 10%. If successful, you can make another roll on the second slot, but still with the 10% chance. This works for the (roughly) 9% TH0, 27% TH1 and 36% TH2 for the above data.

Expected overall drop rate given those rough guesstimates:
TH0: 9% + (9%*9%) = 9.8% (actual: 10%)
TH1: 27% + (27%*27%) = 34.3% (actual: 35%)
TH2: 36% + (36%*36%) = 49.0% (actual: 48%)


It's impossible to figure out if it works the same on the hide drops due to how few sample points there are (I suspect that they have a higher base chance of a second drop; it would be reasonable to have a minimum 10% chance on the second roll), however the above scale makes me rather confident that that is in fact how the multi-drop system works.

Quote:

---

Originally Posted by Motenten

It's impossible to figure out if it works the same on the hide drops due to how few sample points there are (I suspect that they have a higher base chance of a second drop; it would be reasonable to have a minimum 10% chance on the second roll), however the above scale makes me rather confident that that is in fact how the multi-drop system works.

---

I have to say I'm quite intrigued by these results. There are a few points that I find worrisome with this model though, for example the aforementioned higher base chance on certain second drops. Personally I don't think a low cap is very likely, but it is possible. Also, it doesn't explain why we get the results that we did on Colibris, or, for that matter, how TH even increases these drops. According to that, TH would only affect the first item in a multi-drop scenario, the rest would be determined by that new drop rate itself. This should be quite accurate to test on Battering Rams, only getting a large sample size might be a bit bothersome on those.

Quote:

---

Originally Posted by Motenten

TH0 data:
100 total
90 instance (kills with at least 1 meat)
10 2x drops

Raw drop rate: 100/1002 = 10%
Drop rate of total/(kills*slots) = 90/2004 = 4.5% (3.7%-5.5%)
Unique drop rate = 90/1002 = 9.0% (7.4%-10.9%)
Ratio of double to instance: 10/90 = 11.1% (6%-19.4%)

---

I'm guessing the total/kills*slots should be 100/2004? It would raise the percentage, along with the margin of error, so that it would, in fact, overlap with the double-to-instance ratio. However, given the results I still think it's not unlikely (although I'll go over some of the boundary calculations tomorrow, just to be sure).

If this is true, we could test multi-drop behaviour independantly of TH effects (because it wouldn't affect subsequent drops, at least not directly). For TH itself I'd still try to figure it out on mobs with simple loot distribution, and then see how/if we can apply that logic to multi-drops. But this is a nice sample to have, and will be useful for further study.

One more question, are the drops, as listed above, in the order they appeared in in the game?

If +10 from gear alone is the same as +10 from Treasure Hunter I through III plus gear and procs and anything over TH3 has a minuscule effect compared with the first three tiers then /THF with the TH+1 augmented tarutaru sash is 99% as good as an actual THF.

If that is the case then THF really *is* a waste of a party slot.

Quote:

---

Originally Posted by Arcon

I'm guessing the total/kills*slots should be 100/2004? It would raise the percentage, along with the margin of error, so that it would, in fact, overlap with the double-to-instance ratio.

---

Yeah, mistake there. The overlap is pretty miniscule, though: 0.07%.

Edited above post with corrected values.

Quote:

---

Originally Posted by Arcon

There are a few points that I find worrisome with this model though, for example the aforementioned higher base chance on certain second drops. Personally I don't think a low cap is very likely, but it is possible.

---

Actually, the more I think about it the more it makes sense. The nominal drop rate on the hides is something like 1%. If the chance of an additional drop was also 1%, seeing a second drop in the pool would be almost unheard of, never mind 3 at once. Given that the chance of the first item to drop is already low, giving the second drop at least a 10% chance isn't going to dramatically change the overall drop rate, but will at least make it not completely impossible to get multiple drops of an item.

Quote:

---

Originally Posted by Arcon

Also, it doesn't explain why we get the results that we did on Colibris, or, for that matter, how TH even increases these drops.

---

Granted.

Quote:

---

Originally Posted by Arcon

According to that, TH would only affect the first item in a multi-drop scenario, the rest would be determined by that new drop rate itself.

---

That's both correct and confusingly wrong. TH increases the drop rate of the first item. The chance at a second item is based on the drop rate of the first item (with possible lower cap), therefore TH must inherently increase the drop rate of the second item as well. It's just not necessary to extract them as independent elements of the calculation, though you can use the drop rate of the second item to confirm the first.

Quote:

---

Originally Posted by Arcon

This should be quite accurate to test on Battering Rams, only getting a large sample size might be a bit bothersome on those.

---

Got tons of them in Abyssea La Theine :)

Quote:

---

Originally Posted by Arcon

One more question, are the drops, as listed above, in the order they appeared in in the game?

---

No. I can get that from the raw data log, though....

Slot order:

dhalmel meat
dhalmel meat
giant femur
dhalmel hair
papaka grass
dhalmel hide
dhalmel hide
dhalmel hide

Confirmed all combinations. Femur is after meat and before hair; hair is after femur and before grass; grass is after hair and before hide.

There was a confirmed case of non-independent drop slots in Dynamis - Tav pre-patch. Expansion -1 armors could only drop if RotZ non-expansion armors dropped.

Quote:

---

Originally Posted by FrankReynolds

The test required a sample size that big. You are a moron.

Edit: to reiterate. They didn't have to kill 1500 mobs to figure out how much of an increase it was. They killed 1500 mobs just to have reliable proof that there was any increase in drops at all.

---

Which is how it should be, that way a thf is not required, but still provides a benefit.

Also, you whine a lot. Also, I told you so.