What is wrong with reclaim?

[Kenji1134]

This looks like a job for Excel!!!

Here's an experiment for you guys.
Simulate 25,000 random rolls from 0 to 1. If it is less than or equal to 90, call it a success.
Sum up the results of these rolls in sets of 10, and use the frequency function to count how often each set of 10 has a given number of successes.

Here are my general results for sample sizes of 10.
10/10 = 36%
9/10 = 38%
8/10 = 19%
7/10 = 6%
6/10 = 1%
5/10 = 0.04%
Less than 5/10 didnt happen.

Now if we change the sample size from 10 to 25, the results are somewhat different.
25/25 = 8.1%
24/25 = 19.4%
23/25 = 25.4%
22/25 = 22.5%
21/25 = 14.3%
20/25 = 7%
19/25 = 2.5%
18/25 = 0.8%
Under 18/25 did not happen.

So with a smaller sample size, the maximum failure rate was 6/10, or 40% failures in a sample set.
With the larger size, the max failure rate went down to 18/25 or 28% failures in a set.

Ah statistics. Depending on how you present the data, even for a "properly random" source such as Excel, you can see how one person can claim that out of 3 attempts everything failed so they quit. Whereas someone else who (for some unknown reason) does high risk crafting nonstop and does some 100 samples per day would say that the rate is fine. A couple fail here and there but mostly fine.

[ZephyrZ]

This looks like a job for Excel!!!

Here's an experiment for you guys.
Simulate 25,000 random rolls from 0 to 1. If it is less than or equal to 90, call it a success.
Sum up the results of these rolls in sets of 10, and use the frequency function to count how often each set of 10 has a given number of successes.

Here are my general results for sample sizes of 10.
10/10 = 36%
9/10 = 38%
8/10 = 19%
7/10 = 6%
6/10 = 1%
5/10 = 0.04%
Less than 5/10 didnt happen.

Now if we change the sample size from 10 to 25, the results are somewhat different.
25/25 = 8.1%
24/25 = 19.4%
23/25 = 25.4%
22/25 = 22.5%
21/25 = 14.3%
20/25 = 7%
19/25 = 2.5%
18/25 = 0.8%
Under 18/25 did not happen.



So with a smaller sample size, the maximum failure rate was 6/10, or 40% failures in a sample set.
With the larger size, the max failure rate went down to 18/25 or 28% failures in a set.

Ah statistics. Depending on how you present the data, even for a "properly random" source such as Excel, you can see how one person can claim that out of 3 attempts everything failed so they quit. Whereas someone else who (for some unknown reason) does high risk crafting nonstop and does some 100 samples per day would say that the rate is fine. A couple fail here and there but mostly fine.

There is nothing wrong with reclaim. It's a 90% chance, not 10% chance. A sample size of 10 and 25 is pointless and if you're relying on reclaim for 100s of synths then you're doing something wrong. Point being, with a sample size of 100, 1000, 10000 you start seeing it have an average success rate of 90%.

This thread needs to be allowed to die in peace.

[scarebearz]

There is nothing wrong with reclaim. It's a 90% chance, not 10% chance. A sample size of 10 and 25 is pointless and if you're relying on reclaim for 100s of synths then you're doing something wrong. Point being, with a sample size of 100, 1000, 10000 you start seeing it have an average success rate of 90%.

This thread needs to be allowed to die in peace.

The truth is none of us actually know if reclaim is "working as intended" i.e. actually a 90% chance.

It's fine to say that a small sample is poinless, as it is, but where is the large sample to prove it is indeed 90% as stated on the tooltip.

I would say I probably have used reclaim 100 times over the last 2 years. That's like once a week on average.

From my experiance the % is likely to be more than 70%, but I'm not convinced it's actually 90% as stated.


Anyone care to test reclaim 100 times? :P

[ZephyrZ]

The truth is none of us actually know if reclaim is "working as intended" i.e. actually a 90% chance.

It's fine to say that a small sample is poinless, as it is, but where is the large sample to prove it is indeed 90% as stated on the tooltip.

I would say I probably have used reclaim 100 times over the last 2 years. That's like once a week on average.

From my experiance the % is likely to be more than 70%, but I'm not convinced it's actually 90% as stated.


Anyone care to test reclaim 100 times? :P

I tested this a long time ago because i was confused how rng worked, not just for skills but for desynth drops as well. Reclaim, byregot's and hasty touch under SH II, each with a sample size of 250 on junk (Maple Sugar). They average out to 90%, 90% and 80%. Reclaim was the only pain to test, you can test BB multiple times in a single synth, same with hasty.

I also tested materia melding % to realize that it works under same principle as skills and desynth drops. Which is also why you can go on a rng mess and fail 100+ in a row without a meld. It all breaks even.

The biggest confusion is that people thing 90% refers to a probability, it doesn't. It refers to an average. For those that know discrete math, the average in this case would be the expected value.