TH testing: final results

[Esvedium]

Had a "conversation" with some people in Dynamis the other night with them saying that "TH is all placebo lol" and did nothing. Good to see some testing that says otherwise.

[Byrthnoth]

Had a "conversation" with some people in Dynamis the other night with them saying that "TH is all placebo lol" and did nothing. Good to see some testing that says otherwise.

Interpreted as above, these results would mean that TH dramatically and obviously effects things that already have high drop rates (for example, try a run without a THF and look at your single currencies/mob). However, two things:
1) This has not been verified for Rare drops. We don't have a real reason to believe they're different though, I guess.
2) Things like Dynamis AF drops will not be eyeball-ably different. Assuming their drop rate starts at 1% per armor, TH2 bumps it up to ~3% per armor. It'd show up if you collected a large enough sample size, but everything would be within the margin of error for a long time unless you looked at total number of AF dropping.

I'm a fan of the "Adds one roll, adds two rolls" theory. It's simple and meshes well with SE's drop system (aka 10 slots per mob/etc). It doesn't explain why SE chose to put the new job relic armors in different slots though, but SE doesn't do everything with a
http://www.conservenature.org/marine...orpoise_01.jpg and it's too bad.

[Miraun]

It doesn't explain why SE chose to put the new job relic armors in different slots though, but SE doesn't do everything with a porpoise and it's too bad.

Nope, but if it went with my latest example of a variable rolls based on success versus failure with the example of multi-drop Dhamel meat, it would accommodate why some forms of AF appear to have higher drop rates.

Roll on BST, if success, goto Roll on New jobs, if failure, goto DRG
Roll on DRG, if success, goto Roll on New jobs, if failure, goto SMN
.....
New Jobs:
Roll on PUP, if success, goto 100 currency, if failure, goto BLU

This would justify their actions because the observed drop rate on the new armors would be terrible, as it would be subjected to the condition of failing all rolls on the preceding 15 armors... either that or there's a limit on how many imbedded if statements they can use on the computations, and it would be easier just to add a whole new subsection.


Of course, speculation is speculation, and proof would require about 600 million kills of every mob in the game with recording drops, moon phase, day, direction you're facing, and then doing the same 600 million sample size on TH0, TH1, TH2, TH3, and TH4, and again at varying THF levels to make sure there are no additional, unseen ranks of TH that we don't know about. I'm just trying to make sense of what we know in a way that seems logical, I'm actually surprised I got so much support on the first page.

[Byrthnoth]

First off, I'm not really convinced that the "new" relic armor really has a much higher drop rate anymore. I definitely think it did at one time, but either other AF has been brought up to it or it has been brought down. It might have a slightly higher drop rate for reasons I'll explain below.

Secondly, I seriously doubt that it's going to "roll on DRG, roll on WHM, etc." Not to use the overplayed example, but think of it like this:
One /random per slot, slots are, for instance, 1) Old AF, 2) New AF, 3) Single currency, 4) Single Currency, 5) Single Currency, 6) Single Currency, 7) Mob Type-specific crafting item
For the first slot:
1-849 : You get nothing, if you have TH, you can re-/random 1-2 times trying to break 900
850-859 : You get WHM
860-869 : You get DRG
870-879 : You get DRK
etc.

For the second slot:
1-949 : You get nothing, if you have TH, you can re-/random 1-2 times trying to break 900
950-959 : You get COR
.
.
.
980-989 : You get DNC
990-999 : You get SCH

Now, because there are fewer ways to get drops in the second slot, if you have TH you'll be re-randoming more than you would otherwise, so the AF will have a higher drop rate because TH's extra rolls will proc a little more often. For instance:
15% chance of drop rate on first slot, 1% chance of each AF -- TH2 makes it 38.6% chance of a drop on first slot, 2.5% chance of each AF
5% chance of drop on second slot, 1% chance of each AF -- TH2 makes it 14.2% chance of a drop on second slot, 2.8% chance of each AF


Anyway, you wouldn't have to roll for each job to do what you're talking about. And this way is consistent with how we know BCNM drops work (slot has so many options with such-and-such a probability). Consequently, this would mean that times when there's a 100% chance of drop from that slot (like O.Kotes, Deed vs. Cape, etc), TH does nothing for you.

[Miraun]

That would work also... It's entirely dependent on whether or not we observe consistent drop rate across relic armors from one drop slot. If they're consistent, I'm more inclined to go with your proposal, but if there's a slight observational difference for one job over another, I'd stand by mine.

[Byrthnoth]

There's no way to tell that SE coded all jobs to have the same drop rates. All I'm saying is, the simplest way to get the result you describe is to do it with one loot slot as I said.

In order to test this, you would need a high sample size list (years of doing dynamis and meticulously recording even unwanted drops) of, like,
Red Mage : XX drops
Warrior : XX drops
Summoner : XX drops
etc.

which I'm almost completely sure is out there. If a histogram of the drop rates is normal about the mean, then it seems likely it would be my way. If it seems frequency is fairly constant (perhaps in a way you'd predict based on your model), then it may be your way. Overall, it'd be somewhat hard to tell though. I think parsimony is on my side.

[NynJa]

Sumi had an old list like 3-4 years ago tallying up our AF drop rates, I was going to post it, but then it dawned on me that back then we didnt track AF that wound up being trashed.

[Tirgen]

Roll on BST, if success, goto Roll on New jobs, if failure, goto DRG
Roll on DRG, if success, goto Roll on New jobs, if failure, goto SMN
.....
New Jobs:
Roll on PUP, if success, goto 100 currency, if failure, goto BLU

It's very easy to debunk this theory, in a couple ways.
First, the way you have proposed you would never see a Blu, Cor, Pup, Dnc, Sch AF drop alone, it would alway be dropped in conjunction with another relic item. (or you'd never see a 100 drop absent a AF ect). While I can't say for CERTAIN that I've seen one of those items drop alone, I'm almost sure I have, and I suspect somebody that has done a lot more Dynamis than I have probably remembers that happening, and maybe even has a pic. And I'm sure we are ALL certain we've seen a 100 drop with no AF (or shadow pieces....hint hint)

Secondly having dependent drops would make certain AF much more rare. Whichever items are rolled first would be more prevelant, then the items rolled last. Assuming that there is a 1% chance for each af to drop, the first item in line would be rolled at 1% chance, but the next item in line would be rolled at 1% - (Chance that item 1 dropped) I'm no math wiz, so I won;t attempt to number out the actual percentage decreasing, but even assuming it is a very small decrease, repeating it 15 times to get to the last job in line would be a pretty significant bite out of a 1% drop rate

Personally, while I like Byrth's system I think its probably not correct. I would estimate guess that SE did a roll per slot to see if a treasure dropped. So roll a 990+ and any AF drops, then a second roll to determine which AF dropped (lord knows doing 2 randoms isn't going to overload their server with their droprates!)

Doing it that way is much simpler to implement because you add TH1 and now its roll 980+ and a AF drops, then use the same roll as before to determine which one. Add TH2 and now 970+ and a AF drops, again picking the specific one re-using the same code.

Essentially I'd expect it works in a Roll to see if this SLOT gets a drop, then roll to see what that drop is. Which also works with the things we've observed in BCNM/KSNM for example where you always get exactly one of Heart Snatcher, Senji, Expunger, Rampager. You never see 0 of them, and you never see 2+ of them.

This would also explain why treasure hunter doesn't affect chests, or a mob like Mee Deegi -- TH is applied to the SLOT roll, which is already 100% then the code goes right into the specific loot table where the values are not affected. (100% that you'll get Knuckles or Kotes, never no drop or multiple drops)

[Byrthnoth]

Well, there's actually no difference between what you've said and my system Tirg.

Like, say rolling less than 900 gets you no drop, but rolling over 899 gets you a drop and you roll again to find out what it is (100/1000 = 10% chance). Rolling 0-99 on the second roll (100/1000 = 10% chance) gets you one drop, 100-199 gets you a different one, etc. 10% chance * 10% chance = 1% chance.

That's the same as saying rolling 0-899 gets you no drop, 900-909 gets you a specific drop (10/1000 = 1% chance), etc.


I think TH doesn't work on Treasure chests just because it doesn't. Like, if TH worked on Treasure Chests then it would have a pretty big effect on Ultima and Omega body drop rates, or on my personal favorite BCNM's item drop rate. The way I'd explain this is, SE has clarified that a player must be on the hate list and have preformed an action to put Treasure Hunter on something. As far as I know, clicking on a Treasure chest does neither of these things.

[Tirgen]

Well, there's actually no difference between what you've said and my system Tirg.

Like, say rolling less than 900 gets you no drop, but rolling over 899 gets you a drop and you roll again to find out what it is (100/1000 = 10% chance). Rolling 0-99 on the second roll (100/1000 = 10% chance) gets you one drop, 100-199 gets you a different one, etc. 10% chance * 10% chance = 1% chance.

That's the same as saying rolling 0-899 gets you no drop, 900-909 gets you a specific drop (10/1000 = 1% chance), etc.


I think TH doesn't work on Treasure chests just because it doesn't. Like, if TH worked on Treasure Chests then it would have a pretty big effect on Ultima and Omega body drop rates, or on my personal favorite BCNM's item drop rate. The way I'd explain this is, SE has clarified that a player must be on the hate list and have preformed an action to put Treasure Hunter on something. As far as I know, clicking on a Treasure chest does neither of these things.

Despite their failings, I would think that SE has designed thing in such a way that the code for drops is reused everyplace. What I'm saying would accomplish that, as well as leave an easy method for affecting drop rate with TH.


Byakko - (50% drop on Kitty pants)
Slot X Roll 500+ = Drop from loot Table A1234
Table A1234 Roll 0-1000 = Byakko Hadadite

MeeDeegi (100% of Knuckles or Kote)
Slot X Roll 0+ = Drop from Loot Table A2345
Table A2345 Roll 0-950 = Impact Knuckle
950-100 = Kotes

ODS KS99 Chest (100% drop on weapon)
Slot X Roll 0+ = Drop from Loot Table A3456
Table A3456 Roll 0-250 = Heart Snatcher
251-500 = Senji
501-750 = Expunger
751-1000= Rampager

Kindred Demon (1% drop of AF2)
Slot X Roll 990+ Drop from loot table A4567
Slot Y Roll 990+ Drop from loot table A5678
Slot Z Roll 800+ Drop from loot table A6789
Table A4567 0-100 WHM AF
100-200 BLM AF
ect.....

Table A5678 0-200 BLU AF
200-400 COR AF
ect.....

Table A6789 0-300 Byne
300-600 Goldpiece
600-900 Shell
900-930 100 Byne
900-930 100 Goldpiece
900-930 100 Shell


If I was doing it, I would then use TH in the roll for the the Slot -

1) Player rolls random number
2) TH increases that number
3) Modified Number is compared to the threshold

1) Dice Roll! er Slot Roll = 950
2) new roll = Floor(Old roll + (1000 - Threshold) * ((TH1 + TH2 + TH Equip) / 100))

So Kindred Demon Slot X - TH helped only a little bit on an item with a low droprate
955 = Floor(950 + (1000 - 990) * ((20 + 30 + 2)/100) )
3) 955 is less than 990 -> No Drop

Byakko Slot X - TH helped a lot on an item with a 50% drop rate (according to wiki)
1210 = Floor(950 + (1000 - 500) * ((20 + 30 + 2)/100) )
3) 1210 is Greater than 500 -> Drop from Loot Table

MeeDeegi Slot X - TH Did nothing
1470 = Floor(950 + (1000 - 0) * ((20 + 30 + 2)/100) )
3) 1470 is Greater than 0 -> Drop from Loot Table


What I didn't like about your outline was the TH giving you extra rolls. If you do it that way, TH would half to always roughly increase the drop rate exponentially. 2 chances to roll 990+ with TH1. 3 Chances with TH2, then what about equip? Does Thf Knife and ass arms give you roll 4 and 5? It's too big of a jump, and you can't get and extra 'Half a roll' or 42% of a roll

What I like about my method is that is scales with the drop rate. TH gives you .5% better chance on a 1% drop, and 25% better chance on a 50% drop. It also allows SE to adjust Drop rates, add future TH equip/Traits without having to re-balance every loot table. Lowering the Threshold on the slot and/or increasing the TH factor in the equation = more drops, but not all the hassle of changing the AF loot Table to be 974-987 = Blm, 987-1000 = Whm...ect

I like that it fits everyplace - even chests. Slots are 100% so TH would show no effect to the player. Additionally I think you are right, chest have no hate list so there would be no TH bonus to apply anyway

Then again, my method may be a little TOO elegant for SE, so chances are there is a giant hunk of Javascript stuck on each monster to determine drops, half the code is probably broken such that it makes Grah's in RuHmet not change shape....

TL; DR - How they actually did it really doesn't matter, drop rates still suck on just about everything

[Mojo]

The problem Mojo is that with the potential method that I had posted on the first page is that mobs with multiple drops may drop dependencies that we're not aware of... It's a lot more controlled with a single drop item.

I'm aware of this, the data does suggest that mobs with multiple drop loot pools are affected by TH in a different way than those with single drop loot pools, although there is clearly not enough data to be conclusive about exactly how this works. What I suggested earlier (this isn't actually my original idea, I think Shamaya was the first one to suggest it in a thread long ago, dunno though) was that the notion of "no drops" or "some drops" does hold true. All of the data I've collected supports it, and I'm guessing that if Enedin had better recorded his drops, they would support the idea as well.

My take on it is that there are a variety of different types of treasure pools and each pool type is going to be affected by Treasure Hunter in similar yet possibly different ways. Crystals have their own pool and are affected by Signet/Sanction/number of people in your party/level difference between PT members and mobs, but probably not Treasure Hunter (or maybe..?) Synthesis materials have their own pool and are clearly affected by Treasure Hunter, but perhaps not in a very clear way. The chance of the synthesis material pool procing seems to have a set base percentage depending on the mob which goes up by 50% with TH1 and up by 100% with TH2 (or TH1+TH2, whatever) You always get one synthesis item if the loot pool procs, so if the loot pool can only accommodate one item, then that's where Treasure Hunter ends (as seen with the Crawler testing.) This is rare though as most mobs have synthesis item pools that can support more than just one item. In these cases, both TH1 and TH2 have some effect that results in more multiple drops, although there isn't sufficient data to say how this works. This would explain the inconsistencies in Enedin's drop rates (where the total number of items coming off a mob with a multiple drop loot pool goes up by more than the expected 50%.)

To substantiate these claims, review a bit of the data. For Dhalmels there were 1000 kills with TH0 that resulted in a 21.6% frequency of the loot pool dropping and 700 kills with TH2 that resulted in a 52.7% frequency of the loot pool dropping. This is pretty close to the expected base droprate * 2.0 behavior (it's actually 2.4 in this case, which can safely be attributed to inadequate sample size.) The total number of drops I got for TH0 was 249, so a .249 drops per kill ratio. TH2 was 567/700, so 0.810 drops per kill. This means it's unlikely that it's simply item based because a 240% increase in treasure pool procs produced a 324% increase in drops gained (that's relatively 35% off the expected value, which is too large to be chalked up to sample size, despite how low it is.)

Either way, all of the base drop rate data that I gathered supported this, although I wouldn't say there is enough to be conclusive. It still does the best job at explaining what we've seen thus far. There are other types of loot pools that go along with this concept (either or - Defending Ring/Pixe Earring, Abjurations, etc.,) although most of those are hard to observe so it's probably best to keep it simple for now and not speculate about that too much. More data would always help, and I think addressing the following questions while gathering data might be insightful.

1.: Is there a cap on the total number of drops you can obtain from certain mobs?

Dhalmels, for instance, are capable of dropping up to 1 hair, 2 hide, 2 meat, 1 grass, and 1 femur, which is potentially 7 items, although the most I ever saw at once was 4. This could easily be attributed to sample size, although the idea of a certain number of "slots" on a loot pool might be something to watch out for.

I'd also suggest mobs that had decent drop rates on more than one item as you'll see the best results for effort invested with them. Stirges seem good for this, it's too bad that information wasn't recorded.

Also, for people bitching about TH3 and TH4 not being tested. I did do a bit of that testing, although I don't have TH3 or TH4 and I can't really do anything about that. You're welcome to chip in if you want.

Edit: One last thing to suggest is to stop trying to use math on shit we simply don't have enough information about. Cursory qualitative comparisons are ok, and can help lead things in the true direction, but abstract comparisons on specifically what could be happening isn't very helpful unless you have the data to be conclusive about it.

[Byrthnoth]

.216 + .216*.784 + .216*.784*.784 = 51.8%


I believe there's fairly good evidence that monsters do have "slots" that can be filled. Anyone who has killed Jailer of Fortitude can tell you it has one slot for Fortitude Axe (100%), one for Second Virtue (100%), and one for Torque (10-15%).

When you look at monsters, treat their drops like a BCNM. See what items are exclusive. If one drops 2 meat and 2 hides at once, then they'll all in different slots. If you never get Papaka grass and Giant Femur in the pool at once, they likely share a slot, etc.

[Kalisa]

Suggestions for further testing, if anyone is interested in confirming or putting to rest all the theories about whether TH affects rare/ex items....

Rare items: Cardian cards from horutoto ruins. Caveat: You'd have to assume each cardian suit had the same chance of dropping its card (i.e. Ten of Swords has the same drop chance as Ten of Cups). Alternatively, many magic scrolls are rare (Ice Spikes, Stun) but you'd have to target only specific jobs within whole beastmen camps, which would be a pain.
Ex items: Royal Jelly, Crawler's Nest bees. Let a chef or alchemist friend leech the drops.
Rare/Ex items: G1 items, testimonies.

These things have a sufficiently low drop rate that, with an extensive sample size, some nice conclusions could be reached. The mobs are low enough level and abundant enough that there shouldn't be a lot of issue getting the sample sizes you'd need.

If anyone wants to get together and choose some things to focus on, I wouldn't mind contributing a few hundred kills here and there... but I'm not going to go killing the thousands it would take to get significant proof on my own. :3

[Mojo]

.216 + .216*.784 + .216*.784*.784 = 51.8%


I believe there's fairly good evidence that monsters do have "slots" that can be filled. Anyone who has killed Jailer of Fortitude can tell you it has one slot for Fortitude Axe (100%), one for Second Virtue (100%), and one for Torque (10-15%).

When you look at monsters, treat their drops like a BCNM. See what items are exclusive. If one drops 2 meat and 2 hides at once, then they'll all in different slots. If you never get Papaka grass and Giant Femur in the pool at once, they likely share a slot, etc.

You're convoluting loot pools with drop slots, I've described them as very different things. Kirin, for example, could theoretically have different pools for crystals, Kirin's Osode, Kirin's Pole, abjurations, scrolls, and synthesis materials. All of the abjurations that can drop from Kirin can double drop, but I think the most abjurations I've ever seen drop at once is maybe 3 (this could even be 2 or 4, I don't remember.) Thus, that loot pool would only have 2-4 slots (whichever it is.)

[Enedin]

Suggestions for further testing, if anyone is interested in confirming or putting to rest all the theories about whether TH affects rare/ex items....

Rare items: Cardian cards from horutoto ruins. Caveat: You'd have to assume each cardian suit had the same chance of dropping its card (i.e. Ten of Swords has the same drop chance as Ten of Cups). Alternatively, many magic scrolls are rare (Ice Spikes, Stun) but you'd have to target only specific jobs within whole beastmen camps, which would be a pain.
Ex items: Royal Jelly, Crawler's Nest bees. Let a chef or alchemist friend leech the drops.
Rare/Ex items: G1 items, testimonies.

These things have a sufficiently low drop rate that, with an extensive sample size, some nice conclusions could be reached. The mobs are low enough level and abundant enough that there shouldn't be a lot of issue getting the sample sizes you'd need.

If anyone wants to get together and choose some things to focus on, I wouldn't mind contributing a few hundred kills here and there... but I'm not going to go killing the thousands it would take to get significant proof on my own. :3

I've thought about Cardian cards, but you cannot make the assumption that each Cardian's drop has the same chance of happening.

As for other rare items, you say that it's easy getting the sample sizes you need, but I can tell you: finding suitable mobs for testing includes a lot...

- You need unpopular mobs. Can't go kill mobs that aren't there.
- You need a high enough drop rate, or the sample size you need to remove the importance of the error ratio is in the thousands.
- You need enough mobs. 4 in a zone with 16-min respawns isn't enough.
- You can't compare Cardians with each other, just like an Orcish RNG isn't the same as an Orcish BLM.
- The level has to be suitable. Higher level mobs drop more items, but take longer to kill.
- More...

I admire your spirit, but unless you can give us a specific mob to test, it's not happening.

[Kalisa]

I've thought about Cardian cards, but you cannot make the assumption that each Cardian's drop has the same chance of happening.

As for other rare items, you say that it's easy getting the sample sizes you need, but I can tell you: finding suitable mobs for testing includes a lot...

- You need unpopular mobs. Can't go kill mobs that aren't there.
- You need a high enough drop rate, or the sample size you need to remove the importance of the error ratio is in the thousands.
- You need enough mobs. 4 in a zone with 16-min respawns isn't enough.
- You can't compare Cardians with each other, just like an Orcish RNG isn't the same as an Orcish BLM.
- The level has to be suitable. Higher level mobs drop more items, but take longer to kill.
- More...

I admire your spirit, but unless you can give us a specific mob to test, it's not happening.

So did you stop reading my post after I mentioned Cardians, because the other two examples fit all the criteria that you listed as if I wasn't keeping them in mind (I was).... Including the Cardians, since if you go up to the Tens, there are quite a bit more than four. In fact, I listed it the assumption of the same drop rate across multiple types of Cardian as a caveat because I KNEW that was an undesirable assumption to make, so I didn't really need to be told that.

But if you want another option for the Rare-but-not-Exclusive drop... Frayed Arrow, Aht Urhgan Attercop.

[Thorny]

Files a bit more scattered than originally expected, but seems I have about 12000 lesser colibri kills with TH2 and about 9000 with TH0 to parse(estimating by time period on jobs). Should have data posted by sometime tomorrow.

[Enedin]

Files a bit more scattered than originally expected, but seems I have about 12000 lesser colibri kills with TH2 and about 9000 with TH0 to parse(estimating by time period on jobs). Should have data posted by sometime tomorrow.

My original plan was to kill 2500 Gigantobugards in each category, but I never got to that XD (also RMT back then stealing mobs sucks)

Looking forward to these results.

[Masamune]

Just one question: any tests done regarding droprates and THF's Emnity in PT/alliance ?

I personnally experienced better droprates if i go really all out in terms of damage and collabo spam alongwith emnity+ equipements and 0 TAs.... but no parsing and could probably be a placebo effect from some other factors i didnot see ?

I also experienced a lot the opposite if i /sit and just killing blow...

[Enedin]

Just one question: any tests done regarding droprates and THF's Emnity in PT/alliance ?

I personnally experienced better droprates if i go really all out in terms of damage and collabo spam alongwith emnity+ equipements and 0 TAs.... but no parsing and could probably be a placebo effect from some other factors i didnot see ?

I also experienced a lot the opposite if i /sit and just killing blow...

Placebo, eyeball etc.