Goblin Squad Member
|
There are fragments of conversations on Knowledge skills in EE spread around the forum, but I wanted to get a dedicated thread for it to cover basics. I'll dig through and copy things here later (unless someone beats me to it) but at present I'm hoping for two answers:
(1) Are Knowledge skills actively increasing treasure dropped by monsters right now?
(2) Do my ranks in Knowledge help everyone in my party, or just me?
I'm playing a "bard" (enchantment spells, with maxed out Knowledge) and see the Knowledge as a major benefit of my character so others don't have to invest the XP. But I don't want to be out there false advertising...
Thanks!
|
Goblin Squad Member
|
There are fragments of conversations on Knowledge skills in EE spread around the forum, but I wanted to get a dedicated thread for it to cover basics. I'll dig through and copy things here later (unless someone beats me to it) but at present I'm hoping for two answers:
(1) Apparently, yes though the main two that are useful are Local and Nature due to the predominance of humanoids and animals in EE. Also some people claim that taking knowledges to 7 can be detrimental if you are after T1 loot.
(2) My impression is high knowledge is actually more use in a party than solo. When solo you can easily get the loot factor over 100% and get no furhter gain whereas in groups high knowledges are more useful. however that just helps yuo or th ewhole party am unsure.
|
Goblin Squad Member
|
Those are my general thoughts, although some GW confirmation will be reassuring. I've got Local and Geography to rank 6 now since we've got goblins and ogres all around Keeper's Pass, with the rest of the Knowledge skills at 5. I can't say that the treasure we got in our escalation whomping last night seemed to reflect much of an improvement in treasure, but that's too small a sample size to be sure.
|
Goblin Squad Member
|
It is Survival and not Nature, which I have learned the hard way.... Nature will affect the droppings from treemen the day they arrive...
Here is my list, in my order of usefulness
Local - All Humanoids
Survival - Wolfs, Goblin Dogs(rats)
History - Undeads
Geography - Ogres
Planes - Hellhounds
But as low level mobs has some kind of dimishing return, the Ogres are probably the skill that has best Cost/Benefit return over a bit longer time.
Goblins never give anything really exciting, but Ogres always does with high skill.
|
Goblin Squad Member
|
To the best of my knowledge, Knowledges should be contributing to better loot. Survival (animals - wolves), Local (humanoids - goblins, bandits, raiders, knights, cultists), Geography (giants - ogres), Planes (outsiders - hellhounds), and History (undead - skeletons and ghouls) are currently the only ones that have any creatures associated with them, with Local the big winner given our current wealth of humanoids. The other ones will start being useful once we get more creature types.
|
Goblin Squad Member
|
With all of that said, Caldeathe has been keeping detailed loot logs and has seen a noticeable drop in loot quantity since raising his knowledge. It's possible that's just RNG hate, but it's also possible that it's not working as intended.
It's also possible that the system values the drops in a way that's not immediately apparent to us.
For example, the system might consider Broken Goblin Weapons to be more valuable Recipe: Hemp Twine +1. An increase in Knowledge: Local might therefore make it more likely to get Broken Goblin Weapons than that Recipe. This might appear to the user like a decrease in value because "I'm not getting as many Recipes".
I did notice the first few days I played (Scholar, with max Knowledge Skills at the time) that I didn't get a lot of Recipes, but virtually all of them were +3.
|
Goblin Squad Member
|
I think the loot table takes care of the overall value. As far as I can tell it is a 2-step (3-step) process.
Step 1: Checking if you get anything
Step 2: Checking what you get (recipe, weapon, etc.)
Step 3: Checking in subtable the actual loot
Assuming this I made the following table
Recipe (tier 1) 5 4 4
Spell 2 0 0
Weapon 11 9 7
Armor 2 1 3
Tokens 43 40 35
Implements 4 5 9
Salvage 20 13 20
Coin 191 116 ?
Book pages 0 2 3
Total (no copper) 87 74 81
Interestingly book pages only appear later. So it is possible that the loot table changed slightly between the first test and the later tests. If not, then Caldeathe is right and there might be a misplaced sign (minus instead of plus) in the calculation.
Copper has it's own 'subtable' or should I say random number how many you get. Unfortunately you would have to count the times you got copper to add it in the table.
So step 1: approx. 50-60% that you get loot
Step 2: approx. 40% tokens and 20% salvage with the remaining 40% split between various stuff
I hope Lee or Stephen read this and check it out.
|
Goblinworks Game Designer
|
| 2 people marked this as a favorite. |
Knowledge increase should be a flat multiplier on every independent chance the creature has to drop something. If it has 20% chance to drop, and your Knowledge gives you +10%, it now has a 22% chance to drop.
It should not affect any kind of relative chances. If a creature has a 30% chance to roll on its salvage or tokens table, and you get +10%, it now has a 33% chance to roll on its salvage or tokens table, but the relative weights of what's on that table aren't affected.
It does not affect the party, only your own drops (this is part of the reason why you get personal loot that goes directly to your inventory: that loot is yours, was affected by your knowledges, and you're not obligated to share your better loot with people that didn't spend as much XP on knowledges as you did).
There should no longer be any chances that can get to 100% even with maxed out knowledges. I deliberately set them all to under 66% (splitting ones that were over 66% into two independent chances) so even if you're solo you'll never accidentally cap out your loot chance. There's still an advantage to being in a party (a six-member party should, total, be getting around 25% more loot than if they'd killed the same pile of monsters ungrouped).
There shouldn't be anything in the math that creates milestones that could cause weirdness. Rank 7 should differ from Rank 6 in the same way Rank 6 differs from Rank 5: slightly more chance of loot. You guys may be conflating it with the gathering skills (where rank 7 allows you to get Tier 2 stuff, and that DOES mean you can get T2 instead of T1). But the creature loot and knowledges are a completely different system from gathering.
Some things have really low drop chances, particularly when dealing with low-level monsters. For example, the base chance to drop a T1 recipe never exceeds 6.7% (and is 2% for level 1s). That's rolled independently for each creature, and there are no streakbreakers, so my current theory is that any streaks of getting many fewer than expected drops over a big series of kills are balanced out somewhere by someone that got much more than expected drops over a big series of kills. If you kill 1000 creatures with a 2% drop chance, your average result is 20 drops, but that's the peak of the bell curve and there are big swaths of chances where you can get almost none or significantly more than 20.
|
Goblin Squad Member
|
Stephen
What concerned me on Caldeathes numbers was not the recipe - that is insignificant - it was the total as well as the total of tokens.
I think Caldeathe did train up to level 6. This would mean 6% more.
With 189 monsters this equates to 11.3 additional drops. I don't know how many coin drops there have been. But without them we see
13 less, 6 less (after training up). That is 9.5 average - or take coin drops into account - close to the theoretical 11.3 - but less and not more.
Same goes for tokens as well as copper.
So either he really got extremly lucky the first time - or the numbers indicate the opposite that they should be. It is more difficult to move a whole bell curve as a single value. So the total drops and the tokens are the most significant indicators.
So we have a swing of roughly 20 in the wrong direction - or >10% drop rate in the wrong direction for 189 goblins tested. This could approach the 5% significance level that often is used in science to convince yourself that the effect is real.
It isn't a proof yet - but I'm even more concerned now as the numbers seem to fit - just the sign is in the wrong direction. With random numbers it is easy to get a greater/smaller the wrong way and then have a bonus turn into a malus.
|
Goblinworks Game Designer
|
I agree that it's possible there's a bug in the process. Without more data, though, I just can't agree that what you're seeing is clearly a bug rather than just statistical noise. We're keeping an eye on it, but it's not yet to the point of telling the programmers to drop what they're working on to prioritize a drop log.
Goblinworks Executive Founder
|
One set of 189 kills is too small a sample to measure from.
Also, a six percent increase in drop rate is not even remotely the same as increasing the drop rate by 0.06 per kill. If the chance for a recipe was 2% (0.02) at zero knowledge, then it would be (.02*1.06=.0212) 2.12% at knowledge 60.
http://paizo.com/threads/rzs2rhs3?Recipes-droped-ingroup-appear-to-be-broke n#11
An old post that has the actual equation.
Some things have really low drop chances, particularly when dealing with low-level monsters. For example, the base chance to drop a T1 recipe never exceeds 6.7% (and is 2% for level 1s). That's rolled independently for each creature, and there are no streakbreakers, so my current theory is that any streaks of getting many fewer than expected drops over a big series of kills are balanced out somewhere by someone that got much more than expected drops over a big series of kills. If you kill 1000 creatures with a 2% drop chance, your average result is 20 drops, but that's the peak of the bell curve and there are big swaths of chances where you can get almost none or significantly more than 20.
Nihimon and I experienced this exact phenomenon once. I had like 10 recipes from killing mobs in party with him and he had none- despite having a higher knowledge skill.
|
Goblin Squad Member
|
| 1 person marked this as a favorite. |
One set of 189 kills is too small a sample to measure from.
There is statistics to deal with these cases. So lets have a look at it in a statistical way:
I deliberately didn't use the recipes (insignificant - needs too much data) but rather the total number of drops.
87 drops out of 189 kills = drop rate of 46%
Now I need a standard deviation. For simplicity I use the standard deviation for a coin drop which would be 50% but is close enough.
Standard deviation for coin drop = Squareroot of number of kills / 4 = 6.8
So now we have 87 drops plus a standard deviation of 6.8 as expectation
We do 2 more test with a 3% higher expected drop rate (I think he has skill 60).
So our expected value is 87 + 5.67 = 92.67 drops
Experiment 1 is 74 drops, experiment 2 is 81 drops.
This means we underachieve by 2.75 and 1.7 standard deviations. Pooling the last two experiments would boost my numbers and it would make them even more significant.
1.96 standard deviations off is a 95% confidence. I probably would have to do a different statistical test as I compare two experiments with both having a standard deviation.
Bottom line - it is currently 90-95+% certain that something is wrong using Cladeathes data. We probably would have to do it a few more times to be 'certain' - but statistically either Cal was very lucky, something changed between the first and the second try or it cries out something is statistically wrong.
The 10 versus 0 recipe drop also cries out to be looked at. Possibly there is an order in the party that might influence drops to players. Might be good to have a party of 2 or 3 to revisit Cals experiment and kill 189 starter goblins and list the drop rates they achieve.
We know from size of the party how much more they should get.
|
Goblinworks Executive Founder
|
Sorry, but you can't start from one data point and draw a bell curve.
There are advanced statistical methods for this, but those all provide as much support to the conclusion that the first set was high-typical as they do to the conclusion that the knowledge skills aren't working.
And 103% of 87 is about 89, not 92.67. It's 1.03 times the chance to drop, not +.03 chance to drop.
|
Goblin Squad Member
|
Sorry, but you can't start from one data point and draw a bell curve.
You can treat the outcome as binary - any drop yes / any drop no. Unless you can have more then 1 drop from 1 goblin. This depends on the sequence of random variables.
But even then you can bin them into one category as a simplification. And the drop rate was close enough to 50% to use the formula for a coin drop.
There are advanced statistical methods for this, but those all provide as much support to the conclusion that the first set was high-typical as they do to the conclusion that the knowledge skills aren't working.
I'm aware of them - but this was the equivalent of back of the envelop calculation. I'm not doing a scientific piece of work here. I'm just trying to estimate a ballpark figure.
And 103% of 87 is about 89, not 92.67. It's 1.03 times the chance to drop, not +.03 chance to drop.
You are right - my mistake. But again - it doesn't change that much as the attempts with skills are much lower.
Have I proven there is an issue? No
Do the numbers confirm Caldeathes suspicion that there is an issue? They give a good idea that it is possible that there might be an issue.
More data is needed. I will concede if someone shows me data gathered similar to Caldeathe showing the opposite outcome - more loot with more skill. Until then all discussion here is just that - discussion.
We have a founded suspicion that something is wrong and can't tell unless a programmer checks it or we have more data to proof or refute it.
I remember when a lot of people argued against me about the growth rates and that I was wrong there as well. Lee did change them since - to the better.
Hope anyone can provide more data. A good test would also be:
Form a group of 2 - fight goblins. For the sake of it - leave one character just standing next to the goblins and have the other do all the killing. Skill 0, 189 goblins killed - report back.
Rates should be:
a - the same between both characters (off course there will be random variation)
b - a little bit more then 50% each compared to Caldeathes experiment
c - overall more loot as in Caldeathes experiment
If they come in higher then okay. If they also come in lower then we have an issue.
|
Goblin Squad Member
|
To the best of my knowledge, Knowledges should be contributing to better loot. Survival (animals - wolves), Local (humanoids - goblins, bandits, raiders, knights, cultists), Geography (giants - ogres), Planes (outsiders - hellhounds), and History (undead - skeletons and ghouls) are currently the only ones that have any creatures associated with them, with Local the big winner given our current wealth of humanoids. The other ones will start being useful once we get more creature types.
Given the above, what does Arcana, Dungeons, and Nature do?
---------------------------------------------------------------------
Arcana
Dungeons
Geography (giants: ogres)
History (Undead: Skeletons, Ghouls)
Local (humanoids: goblins, bandits, raiders, knights, cultists)
Nature
Planes (outsiders: hellhounds)
Survival (animals: wolves)
|
Goblin Squad Member
|
Survival... Local... Geography... Planes... and History... are currently the only ones that have any creatures associated with them... The other ones will start being useful once we get more creature types.Given the above, what does Arcana, Dungeons, and Nature do?
Stephen's quote answers this question.
|
Goblin Squad Member
|
Stephen's quote answers this question.
I failed my reading comprehension . :(
I am training my adventurer PC in all the knowledge ranks equally, hoping that the missing skills become useful soon and pays of the XP expenditure. At the very least, the contributed a small amount of attribute bonuses...
|
Goblinworks Executive Founder
|
Did we ever get a confirmation that knowledge skills are working as intended? Are people generally satisfied from their own experience that the system is working?
I haven't done A/B tests that would be significant, mostly because I have no intention of counting a few thousand 1st and 2nd level monsters in each of two experimental conditions.