Proving 'Success have a greater chance of giving a skillup'

[Wandor]

Actually Boleslav, Zarazaru is correct. The only way to prove that you are more likely to get a skill up on a success than on a failure is to control every other variable.

I can combine 100 stacks of mino brew and show results where the vast majority of the skill ups are going to come on failures. As a matter of fact, if you pick a recipe with a sufficiently high trivial that has a considerably low success rate, you could feasibly gain ALL of your skill points on failures without a single success. It's just simple probability.

Say for example I decide to make Solstice robes which have a seeminly high failure rate. My tailoring skill is 1 and I have enough supplies to make 1000 attempts. With a little luck with the RNG you should come out with approximately 50 robes (5% chance of success regardless of skill). Regardless of how many skill ups you get on those 1000 attempts, chances are that the vast majority of them are going to come on failures. All this proves is that doing 1000 solstice robes from a skill level of 1 is stupid.

The easiest way to control the success/failure varaible is to give yourself the highest probability of success and failure. From a skill level of 0, make a run of 1000 mino brew combines. The trivial is high enough that you should never have more than a 5% chance of success. Then do another run of 1000 from a skill of 0, this time equiping a geerlok and only doing recipies that are at or near trivial with the geerlok equiped. The test that yeilds the most skill ups would be the winner right?

[Morani]

Then do another run of 1000 from a skill of 0, this time equiping a geerlok and only doing recipies that are at or near trivial with the geerlok equiped. The test that yeilds the most skill ups would be the winner right?

Except that the geerlok seems to work by changing the number plugged into the formula for the skill level (I.E. at skill level 100: the number 100 is used without the geerlok and the number 105 is used with the geerlok.) If the Geerlok where to be a modifier tacked on to the end of that part or somehow affected the number coming from the RNG, it would be a good test.

I can't think of anything that affects the chance of succeeding or failing a combine that isn't affecting the stat or the skill base...

[Silnyil]

Combine Tumpy's tonic
Relatively cheap and higher than Master trivial.
You only need Kiola nuts and water.
So just one recipe you get awful lot of data set.

[bearcaller]

Im not going to take up a ton of extra space and don't really want to cross post (I fear the ogre ) so just take a look at this post (you want the one currently at the bottom:

http://mboards.eqtraders.com/phpBB2/...pic.php?t=7485


Its just a pure simple fact that we should see more skill ups on success than on failure since I know most of us choose the skill up progression with the fewest failures. On items with a better than 50% success rate you should always recieve more skill ups on successes than on failures unless the RNG is feeling especially cruel/fickle/whatnot.

[kiztent]

First: it's a tested hypothesis. Until disproved, it's fact. :P

Science doesn't work that way but your own opinion is you prerogative.

Well, ignoring the political dimension to getting your view out, it does. More accurately, it should. Einstein's feelings about the results of testing the Bell inequality aren't good science.

Of course it requires the mindset that can quickly discard "facts" when they are disproved, but that's roughly the scientific process.

To put it as simply as possible, I'm not really interested in the scientific cause and effect of it. If because of some bug buried in the code that successes lead to more skill ups, I don't care. If there is actually cause and effect put into the code, I don't care. If there is some non separable way to demonstrate that there is not causality, I could care less. You can satisfy your desire for studying the mechanism, I'm interested in real skill ups. Unless you can show how a lack of causality will effect my skill up rate, I'm listening for the roosters :P

Bearcaller:
Go to page 1 of this thread. Read my testing methodology. Look at the results. You can then do one of four things:

1 - demonstrate a fault in the methodology or analysis
2 - develop your own methodology, document it, execute it, show the results
3 - link a statement from sony saying what you are hypothesizing is true
4 - shut up

[Solastar]

Success has NO bearing on skillup

You're wrong, Kiztent (IMHO), and here's why.

Yes you will tend to get more skillups on successes in your experiments.

The reason being is that Zera's work seems to demonstrate that you get more skillups when you are closer to the trivial of the item. Which means, when your skill up rate is good (close to the trivial), you'll get a lot of successes. But when your rate is bad (much lower skill than the trivial), you'll get a lot more failures. So higher proportion of failures when rates are bad vs higher proportion of successes when rate is good skews the resuls. It's the higher skillup rate (due to being close to the trivial) that causes you to have more successes and not vice versa.

If you were to have 100 toons do 10 kiola combines at skill level of below 10, skillups per success or failure would be identical (really bad). Then do 10 combines on 100 toons at 40+ skill and both will have a much better (and identical) skillup rate.

Which at the end of the day means, on a given combine (say a mistle toe sickle at 249 skill) your chance of a skillup is identical, regardless of whether you fail or succeed.

BUT it IS better to skillup on items closer to your skill - because it is THIS that improves your chances of a skillup (regardless of whether you fail or succeed on the combine).

[kiztent]

Re: Success has NO bearing on skillup

The reason being is that Zera's work seems to demonstrate that you get more skillups when you are closer to the trivial of the item. Which means, when your skill up rate is good (close to the trivial), you'll get a lot of successes. But when your rate is bad (much lower skill than the trivial), you'll get a lot more failures. So higher proportion of failures when rates are bad vs higher proportion of successes when rate is good skews the resuls. It's the higher skillup rate (due to being close to the trivial) that causes you to have more successes and not vice versa.

Define close to trivial and how the increase in success scales and I'll consider this. Is it a sliding scale, such that you would notice more skill ups making ol' Tujim's versus Minotaur hero brew (at skill 0 of course)? Is there a magic cutoff that provides a one time increase in skill up percentage when you pass it?

Also notice that on Zera's test runs it took the same number of successes to get to Kiola trivial. It's just that with the sliding skill paths there were fewer failures and therefore less overall combines.

Secondly, and more importantly, how do you use this condition to increase the number of skill ups people get? That is, what scenario is covered by the close to trivial metric that isn't covered by the success makes more likely metric?

[Solastar]

Skillup rates close to trivial

Define close to trivial and how the increase in success scales and I'll consider this. Is it a sliding scale... Is there a magic cutoff...

Well, obviously I can't define it. Despite my grandiose claims, this is just a theory and not proven fact. However, I do feel the skillups close to trivial theory fits yours + Kera's data better than the skillups on success one. My gut feel would be that it's a cut-off point. Say skillup rates improve by 50% when within 10% of trivial or within 10 or 15 points of trivial. (from a programming perspective, I think the latter is simpler, cleaner and more likely)

Secondly, and more importantly, how do you use this condition to increase the number of skill ups people get? That is, what scenario is covered by the close to trivial metric that isn't covered by the success makes more likely metric?

Not sure I follow the fist sentence. But the scenario that gets explained by skillups close to trivial is the one I outlined above. If we run a number of DE mages to skill 10 on Kiolas (and/or Tujim's and Mino Brews) and we see identical skillup rates on fail/success, then it disproves the skillup on success theory (and adds credence to skillup close to trivial - in the absence of any other explanation). I'll run 2 or 3 myself when I get a chance. Hopefully others can do the same. One problem will be that given the low success rate we'll see at such low levels, it will be hard to get a large enough sample of successes to form an opinion.

And identical skillups on mistletoe sickles for success and failure at 245 skill is the other situation which would be a result of the improved skillups close to trivial theory. This could theoretically also be used to test the theories but the expense and much slower rates of skillup at such high levels makes it rather impractical. (not to mention having the smiths/other traders of the right skill level)

In summary - note where I said "in the absence of any other explanation" above. Your data seemed to show prove skillups on success - in the absence of any other explanation. Well, I've proposed an alternate explanation that SEEMS to fit the (limited) data more closely. It will take a lot more experimentation to determine which (if either) is correct).

[Jarak_akn]

For a definative test why not get a large number of people to try the following :

Create a "standard character" (I.e. everyone create the same)
Get one point in a chosen skill
Attempt a combine at a known trivial around 16 (this would give about 45% sucess rate)
See when you get the first skill up is it on sucess or fail.

If we got 10 people to test this ten times we would have 100 data points.
This should give us a reasonable impression. Ideally we would want more like a 1000 data points.

This test would be quite easy once the shared bank space comes in and the success chance of about 50:50 would give us a reasonable spread of results between success and fail.

Are there any other factors that we could control that are not controled by this experiment?

[kiztent]

Re: Skillup rates close to trivial

Define close to trivial and how the increase in success scales and I'll consider this. Is it a sliding scale... Is there a magic cutoff...

Well, obviously I can't define it. Despite my grandiose claims, this is just a theory and not proven fact. However, I do feel the skillups close to trivial theory fits yours + Kera's data better than the skillups on success one. My gut feel would be that it's a cut-off point. Say skillup rates improve by 50% when within 10% of trivial or within 10 or 15 points of trivial. (from a programming perspective, I think the latter is simpler, cleaner and more likely)

At the risk of asking a stupid question how could it possibly fit the facts better? You can't define the metric you are using, yet it is a better fit to experiemntal data, um, because?

Also, until you define the range over which this occurs, it's really untestable, and more importantly, it's unlikely that it's separable from greater success. That is, until you get into the 200ish skill range with a geerlok, items with lower trivial are more likely to yield success.

Secondly, and more importantly, how do you use this condition to increase the number of skill ups people get? That is, what scenario is covered by the close to trivial metric that isn't covered by the success makes more likely metric?

Not sure I follow the fist sentence. But the scenario that gets explained by skillups close to trivial is the one I outlined above. If we run a number of DE mages to skill 10 on Kiolas (and/or Tujim's and Mino Brews) and we see identical skillup rates on fail/success, then it disproves the skillup on success theory (and adds credence to skillup close to trivial - in the absence of any other explanation).

I've done that, and there is a noticable increase in the number of skill ups on success versus number on a failure.

The first sentence is this:
(according to me): If you want to skill up, you have the greatest chance to skill up if you pick the item that is most likely to give you a success. This item is generally (where people care) the item that is closest in trivial to your skill. It also predicts that it is better when skilling from 0 to do intermediate recipes across the skill range to maximize success. It predicts that adding a geerlok in the place of an item with int/wis is a good idea, but doesn't not predict the benefit of swapping a geerlok for an item with int/wis if int/wis are not capped - since the effect of int/wis on skillups is not known.

Your theory predicts that you should work on the lowest non-trivial item available. What other things could you conclude based on your theory? Does a geerlok help (this could be tested but would be annoying)? What things can I do as a result of your prediction to skill up faster?

In summary - note where I said "in the absence of any other explanation" above. Your data seemed to show prove skillups on success - in the absence of any other explanation. Well, I've proposed an alternate explanation that SEEMS to fit the (limited) data more closely. It will take a lot more experimentation to determine which (if either) is correct).

I disagree. It's still not clear to me how this is any improvement on the success theory. Also, it's more specific that the success theory, which gives it less predictive value (assuming of course success is important). In the ranges tested, it's impossible to say one theory is better than another, since they both predict the same results and are qualitative theories not quantitative.

[HoshiAdam]

How about this as an idea:

1. Get a really dumb base character (troll warrior) to test with.
2. Train one point in tailoring
3. Perform the following tests till skill 15:
a. Make silk swatches (a no fail combine)
b. Pelt conversion
c. Cured Silk Armor, or Picnic Baskets (but not the woven mandrake)


Compare skillup rates, based on number of combines needed, making sure to note (as with the previous tests) skillups on success, on failure, and how many success/failures there are for each at each skill level

This test would assume that when checking for a skillup, it first checks to see if an item is successful, then checks to see if you get a skillup. If the chance of a skillup is determined based on being successful (not just being close to the trivial), this order would make sense.

It also assumes that the spider silks won't check for failure to affect the skillup chance. If this is true, the skillup rates for test a and b will be the same. The skillup rates could also be the same if there is an increased chance of skillup when closer to the trivial.

By having a low intelligence/wisdom base character, you decrease the skillup chance, which will increase the number of combines at each skill level.

Any thoughts/ideas about this?

[andyhre]

A Reason to disentangle

There is a good reason to distinguish whether the increased probability is due to being close to the trivial vs. a success, which is the question of whether to equip a geerlock or some other success-increasing tool, vs., say, another int/wis item in your primary. So it is not just two competing explanations without any "actionable" conclusions.

Is there any trivial equal to a no-fail trivial, e.g., if silk combines trivial at 16, is there another tailoring combine that is NOT no-fail, but which also trivials at 16. with enough repeated trials, you should see a difference emerge if the critical driver is a success, and see no difference if the critical driver is proximity to trivial.

[HoshiAdam]

Skinning pelts and the swatches both trivial at 15.

[Modred]

Wrong question

I think we are stating the question incorrectly. It would be much more beneficial to ask, Do skillups come more frequently for items closer to trivial. We are really trying to figure out if working on items that are almost trivial will be better, skill up wise, than working with items that are no where near trivial.

Then the next question would be, as stated several times, is it better to have a geerlook (+5%) than a wis/int item of +X?

[andyhre]

I think the Heady Kiola experiments demonstrate a correlation between staying close to trivial and faster skill ups, which is also a correlation between successes and skill ups. If we can craft an experiment to show that the chance of skilling up on a no-fail is NOT better than on an equivalent failable combine, that would go a long way to eliminating an alternate hypothesis, wouldn't it? If done, right, I think it would put a large dent in the success->higher chance theory. Alternately, if the no-fails do show statistically significantly better chances of skill ups, it puts a dent in the staying close to trivial theory.

Having said that, is it possible to fail a pelt skinning? I think the only times I have ever failed is when *I* accidentally skinned an MQ pelt twice, leaving me with a worthless ruined one instead of the LQ I was seeking. I've done it enough to think it must be no fail, because I fail other 95% combines all the time.