Curse

Nechtan Thaumaturge · Oct 07, 2008, 08:45 AM // 08:45

Where we stand now:

Despite widespread conjecture and debate on the nature of the drop mechanism, no general consensus has been achieved -- at least not on the Guru farming forum. To summarize, for those unfamiliar with proceedings, there are essentially two camps: one maintains that in order to get the maximum number of drops you should only run "one kill at a time" farming builds; the other says that it doesn't matter what sort of build you run, you'll get essentially the same drops regardless. That being the case, you may as well run a build featuring lots of AoE damage and get your run over quickly. For brevity's sake, I'm going to call the former position 1k and the latter AoE.

1k has typically been backed up with a lot of anecdotal evidence -- the "When I run AoE, I get practically no drops" sort of thing. The main problem I have with this is that there never has been any sophisticated sort of investigation done, i.e., nothing with accurate record-keeping over the span of more than one or two runs. You see, given the randomness of drops, anything can happen once. Only over the course of many runs, and we're talking dozens if not hundreds, can you achieve confidence that all the random factors have been weeded out and truth remains. Besides which, I've just had too many 3-drop Raptor runs using Sliver Armor to buy into that theory.

AoE draws confidence from this thread from Skinny Corpse, which would seem to be utterly devastating to the 1k view of things. If you haven't read it, it's worth reviewing the high points at this time. To summarize briefly, however, Skinny Corpse found that if he and his brother zoned-into Fahranur, the First City at the same time, then they would get essentially identical drops regardless of what sort of build they used.

Regardless of where one stands, all acknowledge the looming specter of loot scaling, a game update imposed by ANet on April 19, 2007 which reduced the amount of loot that players could generate through farming. In the words of the developers:

Quote:

People who were advanced solo farmers and who were earning vastly more money through solo farming than through playing the game normally will see the full effect of loot scaling. They will earn less gold and common loot from solo farming than they did in the past. The loot scaling for gold and common loot is not linear with the number of players in the party, and it includes an element of randomness, so while the difference is not easy to quantify, it is by no means a factor of eight. Advanced solo farmers may now earn about twice as much gold and common loot from farming solo as they would if they farmed in a party.

My proposal:

It occurs to me that the debate over drop rates has been mischaracterized. In my opinion, it is not any factor introduced by specific builds that influences drops, but rather kill rate and that alone. This is not a new position; many people have suggested this in the past. However, I hope to offer new insight and proof to support this stance.

First off, many of you will wonder how I could be disregarding the findings of Skinny Corpse's thread mentioned above. First off, realize that there are two distinct points being made there:

The drop system is set up such that two different players can conceivably get the same, or very nearly the same, loot.
There is essentially no difference between the drops obtained using an AoE build and a 1k build.

Point #1 is, to my mind, incontrovertibly proven. However, I feel that point #2 may or may not be true, depending on the circumstances.

Specifically, my position is that, although the two builds Skinny Corpse used feature different skills, there is not a lot of difference between them in terms of speed. Shield of Judgment, while lethal against Undead, results in knockdowns that slow the attack speed of foes. Beyond that, the damage is contingent upon foes actually attacking, so if they stop to cast spells, for example, you do no damage. Also, I suspect the 45 second recharge would delay battles from time to time. Lastly, Zealot's Fire is fairly notorious for causing scattering if not handled with care.

The single kill build, on the other hand, features three fast-casting, quick-recharging skills, two of which don't need to be triggered. It looks to be well-optimized, and I would suspect you could get a kill within the ten seconds it takes for Spoil Victor to recharge. So, even though you're only killing one thing at a time, you are doing this fairly quickly.

Finally, consider the nature of the farm itself. There are relatively few foes per group, typically 3-4, and fairly long hikes between battles. This means that you could never leverage Shield of Judgment's AoE effect much and what gain you might achieve would be diluted by the ten minutes of running that you would have to do over the entire run.

So, I'm saying that the two builds got essentially the same number of drops because they both completed the run in approximately the same amount of time, or, rather, the number of kills each build achieved within some window of time was always nearly the same.

Next, consider these two screenshots:

Here, we have exactly the same build in both cases. Also, in both cases I killed the same number of Raptor Nestlings, 24. The only difference between the two runs in question is the amount of time I delayed between kills -- note the elapsed map times. As you can see, the number of drops increases dramatically in the second screenshot. This illustrates that a single build can yield either a high drop rate or a low drop rate depending how it's played. From this we can infer that the specifics of the build used matter little with regards to drops.

That's the essential point I want to drive home. It doesn't matter what build you use except as that choice affects your kill rate. Many people have tried to tie drop rates to builds, but it's a spurious correlation; the interval between kills is the real predictor of drop rates.

Another thing, it has always been assumed that loot-scaling meant that a solo farmer would see a fraction of the drops that an 8-man team would. Considering the last screenshot, that seems unlikely; an 8-man team would need to get something like 80 drops from the 24 Nestlings. It is possible that the 8-man team would get *better* drops, and thereby make up difference that way, but I'm going to go out on an admittedly shaky limb and suggest something different.

Loot-scaling has more to do with kill rate than with the number of people in your party. In fact, I'm not sure party size matters at all at this point; certainly the difference in drop rates between solo farming and 8-man farming is not greater than about a factor of two, assuming a similar kill rate is maintained.

Consider that idea for a moment, and let's see if I can make it seem reasonable to you. First, what is ANet really fighting with all their farming nerfs? Well, if we are to believe them, and I think we should in this case, they don't want small groups of people to artificially inflate the cost of necessary goods and services as this would tend to ruin the gameplay experience for the larger community. These small groups would effect said price increase by hoarding large sums of gold with which they would be able to make the price on any uncommon resource.

Now, where would they get all this gold? One can always trade various commodities with other players and make money in this fashion, but in the long run supply and demand will coincide, and profit will disappear. In other words, that problem will work itself out. If you and I execute a perfectly fair trade, neither one of us will be the richer, because I can always take what you sold me and in turn sell that to someone else for the same price. Beyond that, the total amount of gold in the Guild Wars universe doesn't change when a trade happens, it just gets redistributed a little.

Rather, it's the *creation* of gold that ANet is concerned with, and that happens everytime you get a drop. Prior to loot-scaling, there was no general cap on drops, and they provided an open-ended facility for getting rich.

Now, for a long time, ANet fought farming with a piecemeal approach. Initially, they singled out players who repeatedly warped out of and back into the same area. Farmers fought back by rotating between areas and lengthening their runs. Then, ANet decided to nerf a couple of the more abusive farming areas, e.g., Griffons -- this was effected by changing the AI and later by adding Kephket and her brood. Farmers simply moved to other creatures and areas. Next, ANet imposed the AoE scattering nerf. Again, farmers adapted by modifying their builds and tactics. And so on, and so forth.

After many failed attempts at balancing, I think that ANet got tired of nickel and dime nerfs and decided to impose a method of limiting drops that no one could get around; enter loot-scaling. It has always been understood that loot-scaling keyed on the number of human players in a party. However, in retrospect, I don't know where this idea came from. ANet has always been circumspect when talking about the details of the drop mechanism, but they have at times alluded to party size when talking about loot-scaling. I think players picked up on this and accepted it without much question.

But think about it. Isn't that a half-solution as well? Players might find efficient eight-man farms and get around loot-scaling altogether. Wouldn't it be better to go all the way to the root of the problem and fundamentally throttle the drop rate in every circumstance? If ANet could do that, their problems would be over. It is true that players would still create gold, but they destroy gold as well.

You see, gold bleeds out of the system in a myriad of ways. When you pay money to create armor, for example, you are forfeiting gold for something which has little value. You can salvage it for materials or sell it to the merchant (at a pretty steep loss), but apart from that it's worthless. Consider lockpicks; they are expensive and destined to crumble. Even if you get a number of uses out of one, the loot has comparatively little value unless you sell it to another player. The entrance fees for the Underworld and the Fissure of Woe are utterly lost once you pay them. Skill purchases, weapon crafting costs, max'ing out the consumption titles... the list goes on. More than anything else, however, people quit the game, often after amassing a fairly good-sized hoard of gold. Some people give away their belongings or sell their account, but how many hold on to their stuff in case they decide to come back one day? All that money may as well not exist.

So, returning to the idea of loot-scaling again, as long as the rate of gold creation doesn't wholly outstrip that of gold destruction then the sum total of gold in the Guild Wars universe remains essentially constant. As long as the amount of gold remains in constant proportion to the sum of all wealth (by which I mean non-monetary belongings), inflation is curtailed. This ensures the fair playing field that ANet desires. Therefore, basing drop rate on kill rate seems quite reasonable to me.

I say all this, because, in point of fact, I have almost no proof to back up my assertion regarding party size. I have run a few trials with encouraging results -- no more than nine drops with an instant-kill build in the Raptor cave, even when four other players were within drop range (thanks, incidentally, to Circus Gimp, Gweneveire Pania, Hogwart The Ugly, and Mordred Jack for helping with this; unfortunately, I messed up the screenshots). The problem is that it's not necessarily easy to put together an 8-man pick-up group to test theories in Rata Sum when most people would rather be farming. Needless to say, this needs some more testing, and I'll have more to say about that later.

Maximizing your profit:

The central question has always been how does one maximize his profit, or more precisely, how does one maximize his profit per unit time (you can make a million gold pieces by wanding Gargoyles outside Ascalon City, it will just take you a very long time).

To help answer that question, I performed a number of Raptor runs and kept track of the number of drops awarded for each. For the initial set of runs, I ran an AoE build and killed everything in one mass. For every other set of runs, I ran a single-kill build and, using a stopwatch with a repeating countdown, timed each kill to occur at a specified interval, in as much as possible. There are five runs per set, except where the kill interval approached 1 minute (in which case I did only three) and where it was relatively small (in which case I did ten or more). I never aggro'ed or killed Rekoff Broodmother or her group. Typically, I killed 23-24 Raptor Nestlings each run. These trials were repeated in Hard Mode. Finally, I averaged the drop percentage over the set and used each as a data point in the following graphs.

The NM drop rate increases quite linearly from 10.9% at zero second kill intervals, i.e. AoE kills, to 87.5% at twenty second kill intervals. Beyond that, the drop rate levels off and stays at essentially 90%. The story is similar for HM, except that the rate of increase is not as great; you have to go out to thirty second kill intervals before you hit the 87.5% drop rate. In other words, slowing down yields less of a gain in HM than in NM.

Don't get confused here and think that you should run a 20 or 30-second build based on this one graph. The problem is that it doesn't take into sufficient account the factor of time. You can get more drops at 20 second intervals than say, five second intervals, but it will take you a good deal longer to do so. If the drops don't improve enough to offset the extra runs that you could have done while you were waiting, it's not going to be worth it.

So, I have done a little math and taken that into account. I had to make some assumptions about the nature of the run being modeled to do this, these being: the amount of time spent not in battle during the run (55s), the amount of time spent warping out of the district to reset the map (25s), and the number of creatures being fought at once (24). The second set of graphs gives you the number of drops you can expect to get in an average hour of farming under these conditions at different kill intervals. These graphs look much different from the first set and from each other. You can see that the region of maximum drops/hr doesn't occur at the highest kill intervals because you can do so few runs in a single hour. By the same token, extremely short kill intervals become much more attractive because you can do quite a few runs in an hour. Lastly, the difference between HM and NM becomes apparent.

However, this also is not the last word because all that I've done so far is maximize drops, not necessarily the gold you would get from them. Here, the concern is that non-loot-scaled (NLS) drops are generally more valuable than loot-scaled (LS) drops and therefore more desirable. NLS drops occur more frequently at shorter kill intervals simply because you are killing more creatures (and loot-scaling doesn't operate on them). So, this factor must be taken into account. Rather than trying to convert all the various drops to their average price, which would be messy, I decided to treat NLS drops as multiple LS drops. The idea goes something like this: some LS drops are worth very little and some are worth a fair amount; regardless of this, we can talk about an average value for an LS drop. For the Raptor run, this would be about 50 gold. Similarly, some NLS drops are worth very little, e.g. most dyes, but some are worth quite a bit, e.g. lockpicks and certain gold items. Again, we can talk about the average value for an NLS drop, and I would ballpark it at around 250g. Perhaps these numbers are inaccurate, but since the values are probably subjective anyway, I'm going to go with them. Next, divide the average NLS drop value by the average LS drop value and get a number, in this case 5. So, that's saying that one NLS drop will generally get you the same amount of money as five LS drops and therefore we can just re-evaluate everything in terms of LS drops.

If that explanation didn't go over well for you, think of it this way. Let's say I've got a bag full of LS drops and you've got a bag full of NLS drops. I can't see what's in your bag, and you can't see what's in mine; however, we agree to trade. I tell you I'll give you one of my drops for one of yours, and you of course tell me no way because you've got gold items, expensive dyes, tomes and such, and I have merchant junk. On the other hand, merchant junk has a certain value as well, so you might make a profit if you can talk me up. So, we haggle and eventually you name an offer which seems fair to both of us. You'll give me one item from your bag for, say, five from mine. To be fair, we agree to grab the items from the other's bag without looking so you don't unload your Orange Dyes on me and I don't give you a handful of Saurian Bones. Sometimes I'll come out ahead and sometimes you will, but the important thing is that you have the expectation that five of my LS drops is roughly equal to one of your NLS drops. Maybe the number is four or eight or six or twenty instead, but whatever it is, that's the NLS equivalency.

So, once we have that number we can talk about "equivalent drops" per hour and that should be proportional to the actual value of our haul in gold pieces. If we can maximize our rate of equivalent drops, we'll maximize our profits also. So, after doing a bit more math, we obtain these two graphs. As you can see, the effect of the NLS drops is to make short runs more worthwhile. In fact, you will maximize equivalent drops in both NM and HM with instant-kill builds.

Having said all this, the analysis I've done only pertains to the Raptor farm and, in fact, only Raptor farms which don't aggro the boss. However, I believe that the data I've collected is representative of drop rates everywhere, once differing NLS drop rates are taken into effect. Different farms will presumably yield different NLS drop rates, and this affects the analysis because the NLS drop rate figures strongly in the performance of instant-kill builds. In general, more NLS drops will occur in HM, due to the possibility of tomes. Fighting bosses also increases the probability of an NLS drop, due to scrolls and green items. Creatures which drop rare materials, e.g. charcoal, silk, ectoplasms, etc., will produce more NLS drops than those that don't. Lastly, NLS drop rates skyrocket during events which feature special items.

Assuming that the drop rate data does apply universally, the five previously mentioned factors (number of foes, time spent not fighting, time spent warping, NLS equivalency, NLS drop rate) dictate the optimal kill rate. If we try to model the NM Destroyer core farm (wait three minutes before fighting anything, add another minute for Sliver Armor recharges and movement under Obsidian Flesh, kill about 55 foes, LS drops are worth relatively more due to the collectible Destroyer cores), the optimal time between kills becomes about 10 seconds.

Rules of thumb:

In general, you want to make short farms shorter and long farms longer. In other words...

Kill rapidly when:

Groups are large
Time between battles is small
Farming in Hard Mode
Non-loot scaled drops are preferred
Non-loot scaled drops are common

Kill slowly when:

Groups are small
Time between battles is large
Farming in Normal Mode
Loot scaled drops are preferred
Non-loot scaled drops are rare

Having said this, the interplay of the several factors is complex and in order to arrive at an accurate answer you should use the Excel sheet I've provided here. To use it, you only need to change the five numbers which correspond to the factors I mentioned earlier. Specifically,

# of foes: simple enough, count the foes you are fighting and enter that number
time out of battle: get a stop watch and time how long you spend in a given run doing everything outside of battle. This includes running to the farming spot, pre-casting enchantments or other skills, aggro, waiting for patrols to arrive, etc. If you start your timer as you warp into the farming area and stop it when you achieve your first kill, this should capture everything.
map reset time: again, use your stop watch and time how long it takes to reset the map, i.e. warp out of and then back into your farming area.
NLS drop rate: in order to determine this exactly, you'll need to count the number of NLS drops you get over several hundred kills. If you just want something quick and dirty, start with 2% and raise it 0.25% for each of the following factors that applies: HM, boss kills, rare material drops.
NLS equivalency: how much more you value a NLS drop as compared to an LS drop; read up about five paragraphs

The times should be entered in seconds; so four minutes is 240, not 4:00 or 400 or 4 or anything else. The NLS drop rate can be entered as a percent without adding the % to the end; so, to input 3% just type 3. Once you've entered the numbers, look at the graph and find the plot that corresponds to the mode that you plan to use, HM or NM. Then, look for the high point on that plot and read the value on the kill interval axis that corresponds to it.

Lastly, I should offer a few warnings. The Excel sheet really only models farms which consist of a single battle. However, you can get an approximate sense for what will work best for lengthier farms by setting both '# of foes' and 'time out of battle' to an average value and making 'warp time' equal to zero. Also, the numbers used in the calculations are still fairly rough estimates; therefore, you should expect some variation as the model gets tweaked in the following weeks.

Conclusion:

This post isn't meant to be the last word on drops, it's meant to be the first. That is to say it's your turn now. Although I have endeavored to be as accurate and correct as possible, I could very well be wrong; so please think critically about what I have written. Having said that, simply posting that you don't believe me doesn't contribute much to the conversation. Reasoning and counter-examples do. Even if you buy into my view of things, however, there's still a lot of ground to cover. I can't say with any certainty that this model is universal, for example, having really only experimented on Raptors. Furthermore, the amount of data I've collected is barely sufficient to draw the conclusions that I have. The concept wants more data and more expertise before it can truly be useful, and I can't give it that all on my own. So please contribute if you are able and so inclined.

Right now, I am most interested in obtaining some data with regards to party size and drop rate. In fact, I'd like to set up an 8-man group via this thread to investigate that, if possible. Please PM me on Guru or in game if you are interested in helping out. It should be simple, although I can't guarantee it will be exciting. You'll need to have a character in Rata Sum, ideally with Hard Mode unlocked. It may be necessary to kill the patrol of adult Raptors that guards the mouth of the cave, but besides that there shouldn't be any fighting for you to do.

Ok, that's that. Thanks for reading!

tmakinen · Oct 07, 2008, 09:16 AM // 09:16

Very good article, having done some related research I completely agree with your conclusions (because I didn't feel like playing with a stopwatch I studied the relation between 'probability of the first killed monster dropping loot' and 'time spent in the instance' and got results very similar to your first graph).

Your reasoning for why the synch entering experiment doesn't say much about loot scaling is also spot on.

I didn't completely get the description of how you handle the LS / NLS issue. Guess that as a scientist I'd like to see the actual equations.

Edit: there's some mislabeling of sections. What you call 'rules of thumb' is really 'conclusions' and your 'conclusions' is actually 'discussion'. Other than that, way to go

Nibbit · Oct 07, 2008, 09:35 AM // 09:35

second post!

On topic: very well done. I was always looking forward to a study like this. 5 out of 5.

Scarn · Oct 07, 2008, 09:58 AM // 09:58

Interesting and well-organized. Good job.

Shasgaliel · Oct 07, 2008, 10:18 AM // 10:18

I think there are also another factors which may be taken into consideration. If you play many characters you have better drops than when playing just one. We do not have a big sample but it seems that drops tend to be bigger when you farm a given area with several characters than repeating it all the time with the same one. Sorry I did not have your patience to make a whole study out of it. Second what comes to my mind is that if you make a break from GW and reduce your playing time the drops also tend to be bigger after your next logging in. Maybe it is just a coincidence but in my opinion it does happen too frequently to be one.

komma · Oct 07, 2008, 10:42 AM // 10:42

Quote:

Originally Posted by Shasgaliel

Second what comes to my mind is that if you make a break from GW and reduce your playing time the drops also tend to be bigger after your next logging in.

this is because your farm code has refreshed....an alternative is doing pvp or helping someone out, that seems to have the same affect as logging off. i/e take a break from farming and do something else for a bit.

Xar Sephtrion · Oct 07, 2008, 11:54 AM // 11:54

Fascinating.

Silverblad3 · Oct 07, 2008, 01:16 PM // 13:16

Great study and post, helps confirm some suspicions I had regarding the kill rate to drop ratio.

Silver

miskav · Oct 07, 2008, 01:32 PM // 13:32

Great post, and it helps to strengthen some of my suspiscions.

natural_Causes · Oct 07, 2008, 01:51 PM // 13:51

Very well done.
I guess this puts to rest the feud of wheather killing faster leads to less drops.

Dreikki · Oct 07, 2008, 03:35 PM // 15:35

You are a god.

Striken7 · Oct 07, 2008, 04:07 PM // 16:07

I think your a bit off on how Loot Scaling works in general. It becomes very easy to see that it IS based on party size when you look at solo farming immediately before and immediately after the LS update.

Before LS, a solo farm run would produce a certain average number of drops. After LS, the same farm run with the same build (same time between kills, same number of foes, same everything else you discussed) would produce significantly less drops (obviously, NLS drops would remain unaffected but the total number still decreased greatly).

Considering all other factors remained unchanged, the only possible conclusions are that Loot Scaling lowered the amount of LS drops regardless of any criteria at all, or that it is based on some factor completely unrelated to the farm itself (skills, speed, foes, etc). Knowing that, and based on evidence collected since then, it does appear that party size has a direct effect on drop rate.

Other then that, I think your work on how kill rate effects drops appears to be spot on.

dasmitchies · Oct 07, 2008, 04:17 PM // 16:17

I think a lot of people are skipping over a common drop rate variable i have found over the last two years. During peak server hours ,when solo farming, the drops have always been more plentyfull and valuable for me. I am EST and I have found Mon. - Fri. form 4pm to 9pm and most times weekends. I recieve better drops than at any other time in the day. (I have played at all times during the day/ week). I have also found non scaled items (festival items, tomes, rare crafting materials, etc.) drop far more frequently whenever there is a lot of people on GW. My hypothesis is the farming code includes a look at how many characters are active on the server when distributing loot as an instance first loads. I can't put my finger on how this works but there is an tangible difference between drop rates during "peak hours" and "off peak". I have also found it doesn't seem to matter if there is a lot of characters playing or just standing around as long as they are logged in. Has anyone else noticed this? I know I ain't nuts here.

jackinthe · Oct 07, 2008, 05:38 PM // 17:38

interested to see what comes of the full party results, can't wait for all the raptor leechers to QQ if it changes nothing

Shursh · Oct 07, 2008, 06:13 PM // 18:13

first off, let me say that this is a very well thought-out analysis on drop rates - congrats!

second, I'd be interested to know a couple things about how you gathered your data:

1. you did all your test runs with either a "1k" or "aoe" build, but were they all done with the same character? the same account?

2. at what time of day and what day of the week did you get your data? were a majority of your drops/data gathered in one sitting?

It's been tossed around that there is a "code" or whatever that limits the number of drops you receive if you repeatedly farm a given area, that the code might be linked to just the one farming character or perhaps your whole account.

If that's true, it might have had an effect on the data you gathered. anyway, just something i'm sure will be brought up sooner or later...

thanks for all your time/effort!!

My Lipgloss is Cool · Oct 07, 2008, 07:14 PM // 19:14

I mostly agree with you, save one point.

I believe nothing changes the droprate of Event Items. I've done some experimenting in both 1k and AoE (I went to extremes. Luxon Assassin Halcyon Job farm. For solo, I killed them with SV and necrosis, and for AoE I balled every sin (except the boss) and killed them all at once with SS)

While I got less drops in my runs, I averaged about the same number of Chocolate Bunnies (It was Easter Weekend) per run.

Also, grats on the massive wall of text

_|\
\|

A Simple Farmer · Oct 07, 2008, 07:18 PM // 19:18

Nice article very well though out.

Question, how big was your sample size for NM and HM raptors? You say, a number of runs. But how many?

Winterclaw · Oct 07, 2008, 08:02 PM // 20:02

Interesting article.

Quote:

Rather, it's the *creation* of gold that ANet is concerned with, and that happens everytime you get a drop

If Anet added more gold-sinks besides too expensive titles then maybe people creating too much wealth wouldn't be a problem. You mentioned UW, if you can average two ectos to drop, then you've recouped your entrance fee by about 8 times. Considering you have to amass so many before you can hand them in, until you do all that wealth is floating around the system.

Also consider "rare" weapon drops that go for high value. Instead of selling them at a merchant where the money can leave the system the money instead changes hands between players and stays in the system. Just imaging how the proph and factions GW economy would have been different if there was a crafter selling prefect crystalline swords for 100K+10ectos.

Fay Vert · Oct 07, 2008, 08:38 PM // 20:38

Q. Why do people farm?
A. To get lots of gold

Q. Why do people want lots of gold?
A. To pay for the gold sinks!

Gold sinks drive the demand for gold, if there were no gold sinks there would be no desire to farm.

Senrath · Oct 07, 2008, 08:52 PM // 20:52

Quote:

Originally Posted by Fay Vert

Q. Why do people farm?
A. To get lots of gold

Q. Why do people want lots of gold?
A. To pay for the gold sinks!

Gold sinks drive the demand for gold, if there were no gold sinks there would be no desire to farm.

Not quite. People want things like rare weapons and minis, which are not gold sinks. Gold sinks remove the gold from circulation, while purchasing rares can only be done through other players, so the gold stays in circulation.