English

OK, just finished trying to make a boostsed set of focus jools. 5420 mats (542 jools netted me 9 focus boosted jools. Add in the 28 oddball boosts that I didn't want and the 4 degrades that means I got 41 boosts. Doing the math:

41 / 542 = 7.5% .... a bit lower than the "advertised 10%".

But more importantly, my tool is dead. With a 250 tool needed for a 250 set, capable tools are going to be in short supply. Now Im not advocating that they be common. But me thinks when ya get ya hands on one, ya oughta be able to at least complete one jewel set.

Suggestions:

1. Increase the durability a lil bit
2. Give us another Advanced Occupation .... ya had one planned to improve crsytal QL..... one which allows ya to improve OP mat QL would be more welcomed as long as it requires an appropriate amount of T & E.
3. Fix the Durability rite

Hmmmm... That does seem to be considerably less durability than the old armilo tools had, by about a factor of two.  I used to be able to get two sets of boosted from an old-style armilo tool.

I agree with your conclusion, however.  One really ought to be able to get a full boosted set from one bit of armilo.

I will add my agreement to this suggestion. It is going to be hard enough to obtain the ql 250 OP mats anyway, so why reduce the lifetime as well.

I am of course assuming that we are not being forced down a route of the biggest, strongest, most aggressive guild will be the only one to enjoy the benefits of ql 250 toolcrafting and ql 250 OP tools and equipment.

Tynee
Squishy little GL of Phaedrea's Tears

Agree Tynee - not sure this game is best when it is pvp focus and homins like the choice - which I think is the strength of the game
*must try and change me picture thing sometime*

Follow up:

With jools outta the way started on 165 focus set.

After 157 crafts have one (1) 165 focus pants.

24 degrades (3 boosted)
5 sap boosts
1 focus boost
2 HP boosts
1 stamina boost

So we have 12 boosts in 157 crafts or 7.6 %

Ummm... FF, that's a very small sample size. I doubt very much that the result is significant. Note, for instance, that you have 24 degrades of which 3 were boosted, a fraction of over 10%. When testing the overall rate of a process affected by an RNG you really need three or four thousand trials before you put a decimal on the percentage.

Your results are suggestive, however, and it would be nice (again) if some dev would either confirm or deny that the boost percentages have changed.

-- B

Bitty:


1. The sample size isn't that small:

542 Jool Crafts + 157 LA Crafts = 699


2. The 3 degrades were included in the calculation:


5 sap + 1 focus + 2 HP +1 Stam + 3 degardes = 12 boosts

12 boosts / 157 crafts = 7.6%



Overall 12 LA boosts + 41 Jool boosts = 53 boosts

53 boosts / 699 crafts = 7.6%

I beg to disagree, FF. You did, indeed, include *all* your results in the count. This is good experimental practice. However, my statement was:
(emphasis added)

The reason for this is based in the mathematics of measurements of events controlled by chance.

When judging the margin of error of a measurement of events generated by random counts, the standard uncertainty is equal to the square root of the number of counts and the 95% expectation level is +/- 3 s.u. The square root of 53 is (approximately) 7. That means that a measured value of 53 could be the measurement of an expectation value (what the theory says) of any number between 74 and 32 at a 95% confidence level. This includes the number 70, which is the expectation value of a 10% success rate for the process in question, and also includes 35 which is the expectation value of a 5% success rate. In other words your current measurement (while *suggestive* of a lower rate than 10%) is not based on enough trials to reliably tell the difference between a 10% rate and a 5% rate.

If you have done 4000 trials the 10% success expectation value is 400 with an s.u. of 20. The expectation value of the 7% success is 280 with an s.u. of 17. A measurement of 370 (9.2%)would be consistent with the former, but not the latter. A measurement of 314 (7.8%) would be consistent with the latter and not the former. Your counting statistics would then be good enough to tell the difference between 10% and 7%. However, even a measurement of 280 cannot yet tell the difference between an 8% and a 6% expectation value, nor would a value of 400 be able to reliably say that the expectation value was not 9% or 11%.

In order to get good (trustable) statistics for a measurement like this, you need a *lot* of trials. The lower the expectation rate of "success" the more trials you need.

All that having been said, a simple answer from the devs as to whether or not they have mucked with the probabilities would save us both a *lot* of time.

-- With respect -- Bittty (who had to deal with this problem at his job all the time).

Bitty,
I agree to your reasoning insofar as Freddy's data do not prove that randomness is broken. Yet, for the unlucky Freddy whose yield of about 700 craftings was -2.6 out of 10 expected there must be a lucky Freddy or whoever with +2.6 out of ten expected, at least on the long range, if randomness is ok.

I know quite well that people seldom if ever complain if chance turns out to their advantage. Yet, before the merge, I felt that exactly that was the case in crafting. Above a success chance of 77% stated in the crafting window, I very seldom suffered degrades or failure, it happened, but much less often than I had expected. It was a striking experience after the fusion that it seems to be the other way round, now.

My first thought was that the devs simply fixed a bug working to our favor and things might be normal, now. Yet my observations contradict such notion.

Roughly a week after the merge, I detected the Dyron overseer as a dapper source, since then, I am crafting two missions twice, sometimes thrice a day to get a couple of dappers. These craftings allow me an observation of outcomes over more than 80 days.

In particular, I am crafting 5 staves and 5 2h swords in the one mission, and 6 shields in the other. Tthere are also bucklers, but as my buckler crafting level developped from 162 to 196 during that time, this crafting is not good for a systematic observation. I had 212 in staves, 214 in 2h sword, and 216 in shields. Roughly after 4 (3) weeks, my according levels rose above 220. I chose to craft swords at the required level of 200, the rest at 210, resulting in success chances of 88%, then 93% for shields and staves, and 92%, then 98% for swords.

Assuming an average of 2x5 swords and staves and 2x6 shields a day during more than 80 days, we are speaking about more than 1600 successfully built weapons and more than 1000 successfully built shields. Beneath that, there was a considerable number of unusable degrades, and also failures. I estimate the degrades and failures at 15-25% of all craftings, making an overall sample of nearly 3000.

Concerning degrades and failures, my impression was that they far exceeded the expectations. Until the success chance for swords rose to 98% (about 3 weeks ago), I had no single series of 5 without a failure or degrade, even now, one or two, making 6 or 7 craftings necessary, are quite common. Until now, I had no single occurrence of crafting 5 staves or 6 shields (chance first 88%, then 92%) in a flawless series. On the other hand, 5 or 6 unsuccessful craftings in a row well occurred though seldom.

I know that these data, with gliding success chance, with results not correctly written down etc. do not allow a rigorous analysis, and I regret to have no better. But it strengthens the suspicion that something is wrong with random algorithm in crafting.

To me, it looks a bit like a normal distribution where the right side of the graph is cut. I could program such an outcome, it is not hard, and may as well occur unintentionally by some side effect of a bug.

That means that tough luck still happens as everywhere, normal 100% success, too, but the lucky outcomes are missing. That is what I felt when I read Freddy's articles.

I agree that a word of the Devs would be good.

This is a knowledgeable and entertaining discussion. It might be lame of me to add another viewpoint with the summary that the devs should have the final say, but here it is: Independent of all that has been said, the server merge possibly also included a migration: I never checked the Ryzom shard IP address, so cannot tell if indeed it has moved. But if it has, it has likely been migrated to a newer machine. Changing the machine, or changing the operating system (even just versions or configurations), may change the way random numbers are generated.

Below is for those who may wonder why this is relevant.

Ryzom depends on sufficient random number generation to also warrant special algorithms to deal with their creation. For anyone interested "Mersenne Twister" (1997) and subsequently "WELL" (2006) are interesting starting points for reading. I have no idea what the devs chose to use (maybe even plain rand() by the OS), but it may be part of the problem. Possibly a problem in the way the numbers are generated surfaces now after the merge only because of a hardware change.

Here's an example of the "randomness" problem: If you linearly write down random numbers you get from a generator, they may appear to be totally random. And maybe this is proven by some statistical analysis of those numbers. Now, if you take the same sequence of random numbers but write them down not as a linear sequence, but as a matrix. For example, if it's about 100 random numbers, write them down in a ten by ten square. Suddenly, the random numbers could appear to no longer be so random, looking at just the columns, or looking at just one specific diagonal. While hard to write down and observe, this can be extended to 3 or even more dimensions.

Why is this relevant? Software such as Ryzom does not dole out random numbers to just one user/client. If there is a constant number of clients connected, and the server generates one random number for each client in turn, then the next, then the next, and so on (round robin), then we suddenly must have random numbers which are random (uniformly distributed) not just in one dimension, but in two. Of course, in a game, lots of clients connect and disconnect all the time and one client may require more random numbers than others (crafting versus exploring). But indeed, uniformly distributed random numbers across a higher number of dimensions is important, and ultimately led to the development of Mersenne Twister. Generally speaking, the higher the number of "random" dimensions, the better. WELL was developed to fix a couple of other important attributes, but that's for another story ;-)

I draw my hat before people like Bitty for the statistical work which I never liked, and also Freddy and Daomei for the experiments. Thanks for your enlightenment.

Bitty:

I didn't say the sample, wasn't small. I said it wasn't THAT small. Yes it is possible that the statistical deviation makes the 10% average still possible but with each increase in number of crafts, that chance gets smaller & smaller. And it's eerily odd that the LA % and the Jools % were the same.= %

Also of significance, as Daomi pointed out, we haven't heard of a Lucky Freddy. Peeps I have talked to are reporting the same 7%, or at least less than 10%, and even more oddly, everyone seems to be getting an overabundance of sap boosts. The commonality of experience "seems" just too odd here to be a statistical aberration. Time will tell but I imagine an election newscaster, would have the confidence at this point to "call this one". Yes, there's the Dewey / Truman and Gore in Florida thing in the back of the newscaster's mind but just how often does that happen ?

As pointed out by Gilga here, the random number thing "has" changed:

http://app.ryzom.com/app_forum/index.php?page=topic/view/15434/62

If we drop the decimal and call it 8%, we still need 25% more to get to the "given" number. I have had my issues with RNG in game before.

I went 0 for 66 on the "/random 100" thing in Dante without thinking it was broken and then managed 3 wins outta the next 10. While that was still only 4% to the expected 11%, I wasn't about to jump to a to a conclusion

I remember that there was a similar discussion back when we started using scrolls. However, I think that few would argue today that the 99% (85 + 14) success rate with a QL70 scroll was real.

Like you, I also do math "for a living" and while I'm not quite ready to "bet the ranch", the commonality of experience makes it worth my while to continue looking at it.

Heh -- Looking at these last few posts it would seem that FF, Daomei, Irfidel and I are all in violent agreement even though we are coming at it from several different directions. :)

I would say that I seem to be (on the average) "Lucky Freddy". My results (anecdotal because not written down) have almost always come out close to the ones I expect and I do get runs of success at 85%. The one time I did a commission of two sets of boosted LA for Beeficus and Beefie I actually completed both in well under the theoretical number of mats based on simple expectation value. All of this, of course, being pre-merge. Post merge I seem to be getting about the right number or a bit more of successes when overcrafting at the 5% level.

As for boosts, well I don't own an Armilo mine any more so I don't have it to play with post merger. I do know that when I was doing boosted armor, however, when I was doing HA with hp boost almost all the boosts I got were HP, while if I were making jewels or LA with HP on them I got a much more random distribution of types of boost.

OK, today's results

232 crafts have one (1) 165 focusboots.

40 degrades (boosted ones included in Nos. below)
5 sap boosts
1 focus boost
7 HP boosts
4 stamina boost

17 boosts / 232 = 7.33 %

===============================================
After 389 LA crafts have one (2) 165 focus pieces (0.5 %)

64 degrades (16.5% ~ 16.7% anticipated)
29 boosts (7.5% ~ 10% anticipated)

===============================================
Overall 29 LA boosts + 41 Jool boosts = 80 boosts

80 boosts / 1088 crafts = 7.4%

What would interest me: When using scrollmaker products q60 or 70 (theoretically boosting success chance to 96-98%), is there an effect now? Before the fusion, scrolls had no effect on master level, at least not according to a couple of tests I ran.

Before merge I know the Q70's didnt work .....certainly not the 99% "advertised". I had much better luck with Q50s.