The A.M.A.N. Trove BC

Pulled logs from my mains, 1690 total chests opened.

46 loud thud
274 thud
1370 noise

So, we can roughly assume:

1370/1690 * 50k = 40,533g
274/1690 * 300k = 48,639g
46/1690 * 20m = 544,379g

Thus, a chest is worth ~633,551g using YOUR values.

Now, let's look at your strategy.

Pt1: Open after 3 thuds.

3 Thuds, 2 noises (5 chests) : 1M value + 633,551 from final chest.

80% chance of success * 633,551g = +506,841g
20% chance of fail * -1,633,551g = -326,710g

In this case, you stand to make 180,131g on average by opening a 6th chest. Guess your strategy made the wrong choice.

Pt2: Open after 7 chests.

6 noises, 1 thud (7 chests) : 600k value + 633,551 from final chest.

2/3 chance of success * 633,551 = 422,367g
1/3 chance of failure * -1,233,551 = -411,184g

In this case, you stand to make 11,183g on average by opening an 8th chest. Guess your strategy was wrong again. In the event of 7 noises, it is 94,517g.

Pt3: Open after 1 loud thud.

This is correct, but everyone already knows it. Even if it's the first chest:

8/9 chance of success * 633,551 = +563,156g
1/9 chance of fail * 20,633,551 = -2,292,617g

Lose 1,729,461g on average by opening that second chest.


I AM NOT CLAIMING THESE NUMBERS ARE ACCURATE. I AM USING SORALIN'S VALUES PER CHEST TO ILLUSTRATE THAT HIS STRATEGY DOESN'T EVEN MATCH HIS OWN NUMBERS.

The truth is that you'd need tens of thousands of runs correctly logged to actually know with any certainty. Obviously a chest type doesn't average to 50k, 300k, 20m on the nose. I don't think they're even close. Further, we have exceedingly little information about the actual distribution within the final chest. But, fact is, we don't have the data and soralin's conclusions are drawn off emotion and feeling, not numbers.

But...but, the rainbows and unicorns demand their fair share representation against the facts too!

Based on your understanding of the algorithm, your assertion would be correct.

I however do not believe the algorithm to work the way you assume it does and have modified my strategy to fit my understanding of the chests algorithm.

Perhaps 2 pages ago you could've started off with "Well based on my understanding of how I think the chests work, I think your strategy sounds sub-optimal for x y z reasons and these numbers" we could've had a great discussion and batted ideas back and forth like two rational adults.

And you dont need thousands of runs. The loot pool is small, its a *very* safe assumption to assume the value of most money drops off the Mars Orb to be worth 200mil, and its pretty safe to assume a 'best fit' strategy to the constraints that the loot pool is evenly distributed.

It might not be, but even if it wasnt, a weighted distribution tables best fit strategy would not be terribly different from an even table distribution strategy.

So you can estimate a 'close enough' plan.

Intelligent people will first resort to "Hmm, you have contradicted my perception, lets talk about this"

Children resort to "That doesn't align with my thoughts, so you MUST be an idiot and couldn't POSSIBLY know something I don't!"

Absolutely 100% I will just buy it next time I see a shout, Lesson learned. I just thought I'd at least get something nice from over 100 orbs in total over the past few months.

You're just playing word games at this point. You still have yet to provide anything of substance, and it's quite obvious to me (and likely others) that you don't have anything such to present.

You're the one who insisted there's no pattern, so how exactly does your algorithm differ from straight probability calculation?

The most effective system/method/strategy is sell your orbs and buy the thing you want with the profit. It works 100% of the time. (buy the thing you want and sell orbs to payback the debt*)

Buying hundreds of orbs and selling the drops you likely come out ahead, but it's still a gamble!

I feel no real reason to help out someone being an *** about all of this. This could've been a cordial and productive convo. You were the one who made the choice to make it anything but.

You keep dropping that you're too good to help me (I still don't remember asking for or wanting your help), but the truth is that whatever elevated understanding you claim to have can't be taken seriously if you just say things along the lines of 'I know, but I won't tell you!'. If you actually know something I/the forums don't, why not say it and rub it in? Then you wouldn't need to keep posturing, and we could all learn something.

Nah.

Random is random, and unless you enjoy the feeling of winning the lottery (aka never happens), just cut your losses, set a base price, and sell the rest of your orbs for life. it practically pays for itself that way, you just put the money out up front

Solid.

Based on Comeatmebro's approach to the algorithm, I whipped up a very rudimentary C# program that will brute force trials with best fit selection strategy.

Its expected you fill out the constants with values matching your own data and etc, then it will run the trials optimizing for maximum profitability.

Then finish off by printing out the net average gil made per run.



As you can see, based on my estimated values of Heavy Thuds and etc, the profitability margin is actually incredibly slim.

Link to the program's source:
https://github.com/SteffenBlake/AMANCalculator

I don't even know what the drama is but I assume it's some random number.

FFXIAH is the king of random number arguments.

I did 10 million AMAN chest and my conclusion is *punches numbers on calculator* it's a waste of time !

Nice to see your contributions to threads are as vapid as ever. If you have nothing to contribute, go somewhere else.

Yea please post more random numbers to prove how good at opening AMAN chests you are.

If you have an issue with the best fit algorithm, feel free to open an issue on the github, since it literally has a feature for that.

If you're statements summarize to "I don't understand therefor it must be wrong" then you're not contributing and can leave.

Posting random numbers doesn't prove anything. It just means you like to post numbers to argue.

So the second option.

If anyone who actually wants to contribute to the convo is interested in participating, I'll field questions about the program or whatever.

But I feel like I commented it well enough to lay out how it works.

Where did you get the distributions from?

Also, I might fix this to learn an optimal policy using Monte Carlo Tree Search or something.

Oh man is he back at it? Is his subscriber count down?

https://www.ffxiah.com/forum/topic/52655/the-aman-trove-bc/47/#3428053

MCTS is only super useful on non-deterministic strategies.

Since AMAN Trove is assumed to be deterministic, as your 'opponent' is a predetermined chance roll, then MCTS isn't very helpful.

Specifically, once you cancel out your pieces of the MCTS algorithm that work out to contributing 0, you'll find the whole MCTS algorithm is removed and all the remains is just the statistical formula you started with :)

MCTS works for a game like, say, tic-tac-toe, where your 'opponent' has the capability to 'choose' and you wish to min/max your choice based on the assumption your opponent also plays the the best of their ability.

Which is why a naive MCTS approach loses to 'random' players that make unpredictable moves.

But it still could prove to be a fun thought experiment to allow you to see in practice how deterministic strategies cause the MCTS algorithm to simplify out, which does help in fully grok'ing how MCTS functions holistically.

Do we know that the placement is deterministic w/r/t a prior distribution? idk if the chest contents are decided as soon as the instance loads or reactively as you open chests. I assumed that the chest contents responded to chests opened in the same run, e.g. the system prevents you from getting all loud thuds. We know that each instance has a mimic but beyond that I have no idea. I guess the pre-loading placement theory is assumed so operating with expected values is better

Exactly, thats what I prefaced in the README file on the github :)

readmes are for squares

I'm new to the forums and I'm going to bring in some positive ideas for AMAN Trove:

Make it so that there is a rare chance that some Dyna-D stuff drops. Like if you're on THF, some a THF Shard can drop.

We'll get right on that, great idea.

You mean Volte.

There is very little reason to want Void/Shard from Trove, it's money stuff.