[WCA Regulations 2014] Scramble Filtering (new poll)

[tseitsei]

We're definitely going to have arbitrary limits, but if we do filter things beyond 1 move, I'd like to at least have consistent reasoning across puzzles, as opposed to arguments about whether an N-move scramble feels easy or can be solved by a non-cuber or whatever. One possible way is as follows:
- Choose a constant k, much less than 1.
- For a given puzzle, let P be the number of positions.
- Let n be the greatest integer so that the number of positions solvable in at most n moves is no greater than kP.
- If computationally feasible, forbid any positions that are solvable in at most n moves.

For instance, if we let k = 1/1000, we forbid the following: 2x2x2 positions solvable in <=4 moves; Pyraminx positions solvable in <=3 moves (or <=5 moves if you include tips); Skewb positions solvable in <=4 moves; Square-1 positions solvable in <=6 twists; 3x3x3 positions solvable in <=14 moves. Unless I made a mistake, of course.

I think that if we are gonna do scramble filtering (which I don't really want) then something like this is the correct way of doing it.
Not just throwing some random numbers out of nowhere to different puzzles...

 

[mark49152]

I'm neither a competitor nor delegate so my opinion is a view from the sidelines, but for anyone who's interested in a sideline view, I find this whole discussion quite bemusing. Surely it's not the randomness that matters, it's the quality of the test. The whole point of random scrambles is to try to ensure a uniform test across competitors and competitions. Randomisation is only a means to an end, not the objective itself. If you allow lucky scrambles that degrade the test to the point that less skill is required to succeed, just because randomization can occasionally produce such scrambles, doesn't that undermine the competition?

 

[Kirjava]

one or two moves also just feels "not scrambled" to me

The only justification I see for this line of thinking is "because it does" though.

I understand why you think that, I just think you're wrong.

Maybe now you or one of the others who think we shouldn't even filter the solved position can enlighten me about that, cause that's something I don't understand. Do you consider a solved cube scrambled? Or do you think we don't need to scramble?

Nah, I can't. I think we should use actually scrambled states for scrambles.

People think that by introducing this filter they are solving a problem. Most people think that this is not a problem in the first place.

However, the problem they are 'fixing' still exists, it just behaves slightly differently. You just need to look at the 2x2x2 single WR list to see the quirk of a single 4 move scramble happening.

 

[Stefan]

Well, that might be a bit misleading, especially to someone who hasn't read discussion regarding 1-move scrambles. But I'm fine if someone wants to change my vote to option 2.

How would that be misleading? The question is "Should the scrambling procedure include filtering certain positions?" and if you're for 0/1 moves filtering (but not more), option 2 is clearly the correct choice. Saying "no", on the other hand, is not just misleading but also simply wrong. It would be wrong even if you were only for filtering the solved position.

We're definitely going to have arbitrary limits, but if we do filter things beyond 1 move, I'd like to at least have consistent reasoning across puzzles, as opposed to arguments about whether an N-move scramble feels easy or can be solved by a non-cuber or whatever. One possible way is as follows:
- Choose a constant k, much less than 1.

How do you choose that k, though? Is that not guided by whether the resulting scrambles feel easy or can be solved by a non-cuber or whatever?

For instance, if we let k = 1/1000, we forbid the following: 2x2x2 positions solvable in <=4 moves; Pyraminx positions solvable in <=3 moves (or <=5 moves if you include tips); Skewb positions solvable in <=4 moves; Square-1 positions solvable in <=6 twists; 3x3x3 positions solvable in <=14 moves. Unless I made a mistake, of course.

I get the same numbers except for Pyraminx with tips, where I get <=6:
1+8+48+288+1728+9896+51808 < (1/1000)*933120*3^4

 

[Methuselah96]

2 points:

- it doesnt matter how you feel about a scramble. A scramble is no more than a random position of a cube in my opinion. This is the point where a lot of us just have a different opinion. Imho: If you dont like the fact that one of the random positions is 1 move (which is only considered solved because of our regs) then why bother with a sport-hobby-etc of where luck is a clear possible factor. A factor which, is already filtered by averages of 5.

- yes lets use non-cubers insight in cubes to determine what is a scramble or not...

What if one of the scrambles is a solved puzzle? Is that a possible scramble by your standards?

I don't agree that how we've handled scrambles in the past should set a precedent. If skewb were to be introduced with a minimum scramble length of 9, and then we realized that was stupid, I don't think we should be stuck and never be allowed to lower the minimum. Rather, those cubers who competed under scramble filtering and had a low-move scramble removed were just unlucky that the WCA was handling things this way at that time (similar to cubers competing under the Rubik's-brand-only rule).

Looking at the long-term, we can't never change the way things are because of a ruling at one competition, it's simply unfortunate to cubers at that competition. Following this reasoning, I think that neither the case of throwing out a 2-move scramble nor the case of allowing a 4-move scramble for 2x2 set a precedent. I'm in favour of allowing all non-solved states to appear (where 1 move off is solved). This seems to be the most non-arbitrary course of action.

I also fear that we could be stuck with too low of a filter because of past filters. But I think he's saying the opposite of what I'm thinking. He's talking about not lowering the filter. I'm talking about being able to raise it, if need be, but I don't know how that would work fairly.

 

[Stefan]

it doesnt matter how you feel about a scramble. A scramble is no more than a random position of a cube in my opinion.

it doesnt matter how you opine about a scramble

(don't see how "opinion" is better than "feeling")

If you dont like the fact that one of the random positions is 1 move (which is only considered solved because of our regs) then why bother with a sport-hobby-etc of where luck is a clear possible factor.

I don't know the name of that fallacy, but I'm sure it has one.

yes lets use non-cubers insight in cubes to determine what is a scramble or not...

Indeed. For them it's really still a puzzle, they're not biased by too much knowledge. I consider that valuable and meaningful.

(yes, I do understand you disagree)

 

[Erik]

I do think it is a really bad idea to take the opinion of non-cubers into account. I base my concept of a "scramble" on a random position of the cube -which is a logical way-, rather than on a feeling. It surprises me a beta/numbers guy like you doesnt do this as well.

 

[Stefan]

one or two moves also just feels "not scrambled" to me.

The only justification I see for this line of thinking is "because it does" though.

Then you didn't see my two sentences right after that:

one or two moves also just feels "not scrambled" to me. Even non-cubers can solve that easily. And wouldn't give it to us when asked to scramble.

And another way I thought of it was with eggs (the only thing besides cubing that comes to my mind when thinking "scrambled"). If I order a scrambled egg, I neither want a raw one (zero moves) nor a sunny-side-up (one move), nor a "flipped sunny-side-up" (don't know a name for that) or a cut sunny-side-up (or whatever else I can imagine being two moves). None of those are acceptable when I want it scrambled.

I know it's not a perfect analogy, but the point is mainly that just because it's not solved/raw anymore, doesn't mean it's "scrambled" and sufficient.

Nah, I can't. I think we should use actually scrambled states for scrambles.

So you're for filtering the solved position? Then why did you vote "No" when asked "Should the scrambling procedure include filtering certain positions?"?

People think that by introducing this filter they are solving a problem. Most people think that this is not a problem in the first place.

Don't know about the "most people" assertion, but I've seen enough people unhappy with at least the current 2x2 and square-1 WRs to think that if they were beaten with two-moves scrambles, that "most people" assertion might become false.

 

[Kirjava]

Then you didn't see my two sentences right after that:

I did and i don't accept them as justification.

So you're for filtering the solved position? Then why did you vote "No" when asked "Should the scrambling procedure include filtering certain positions?"?

I took the question to be asking if we should be filtering scrambles. I don't consider the solved position to be a scramble.

Don't know about the "most people" assertion, but I've seen enough people unhappy with at least the current 2x2 and square-1 WRs to think that if they were beaten with two-moves scrambles, that "most people" assertion might become false.

Either way it's not solving the problem.

There's no objective reason to set it anywhere other than two. You claimed we had precedent for removing a two move scramble in 2x2x2, yet the limit for Pyra is 7 after a 6 move scramble already appearing in competition? This is just inconsistent and if the poll is anything to go by this issue needs a serious look.

 

[Stefan]

I did and i don't accept them as justification.

It's distinct from and better than "because it does", and you called that a justification.

Anyway, you might not consider it a justification, but please don't try to make it look like I didn't provide anything.

I took the question to be asking if we should be filtering scrambles. I don't consider the solved position to be a scramble.

The poll question explicitly asks about "positions", not "scrambles". I wonder how many of the other "No" voters made the same mistake.

There's no objective reason to set it anywhere other than two. You claimed we had precedent for removing a two move scramble in 2x2x2, yet the limit for Pyra is 7 after a 6 move scramble already appearing in competition? This is just inconsistent and if the poll is anything to go by this issue needs a serious look.

Earlier today in the "official discussion" about it, Sebastien wrote

"People also pointed out that the Pyraminx Single WR was 6 moves which seems to be true while I thought it was 7 moves before. I agree that we should change Skewb and Pyraminx from 7 to 6."

I base my concept of a "scramble" on a random position of the cube -which is a logical way-, rather than on a feeling

It's only a still only "feeling" because it hasn't been measured yet.

 

[KongShou]

It's only a still only "feeling" because it hasn't been measured yet.

What are you trying to say? There is no verified scrambled state? We should filter one or two move scrambles because they dont feel scrambled? Man i respect you most of the time but please present your argument with some justification.

 

[angham]

Lel, my official 3.61 pyra would never have happened

 

[qqwref]

How would that be misleading? The question is "Should the scrambling procedure include filtering certain positions?" and if you're for 0/1 moves filtering (but not more), option 2 is clearly the correct choice. Saying "no", on the other hand, is not just misleading but also simply wrong. It would be wrong even if you were only for filtering the solved position.

Relying on technicalities to get more people to vote "yes"? That's just low.

How do you choose that k, though? Is that not guided by whether the resulting scrambles feel easy or can be solved by a non-cuber or whatever?

No, I don't care about whether the resulting scrambles feel easy or not - a sune plus AUF on the 2x2x2 feels easy, but that's already 8 moves, and no reasonable filtering method should remove that. k pretty much provides an upper limit on what proportion of positions we will exclude (so k=1/1000 has us exclude no more than 1/1000 of the positions for a puzzle) and thus guarantees the maximum possible effect our filtering has on the distribution. We should probably either choose a nice round number, or choose a number that gives us a desired amount of exclusion on one puzzle - as an example, if we feel like 4-move 2x2x2 scrambles should still be allowed but no easier scrambles should show up, we could choose a k that falls in this range, which is about .000104 to .000607. At least, once we have made one arbitrary choice, the rest of the limits follow naturally.

 

[Stefan]

What are you trying to say? There is no verified scrambled state? We should filter one or two move scrambles because they dont feel scrambled? Man i respect you most of the time but please present your argument with some justification.

The "feeling" part is where I *speculate* about what non-cubers would find easy rather than having measured it. In my opinion, if the majority of average-intelligence average-health non-cubers can solve the majority of N-moves-scrambles easily, then N-moves-scrambles aren't good enough. And that can be measured (not exhaustively, of course, but if you do enough tests, you can probably find the answer with enough certainty).

 

[Kirjava]

Anyway, you might not consider it a justification, but please don't try to make it look like I didn't provide anything.

I'm not trying to do that, I'm merely saying that I think what you're saying doesn't hold much ground.

The poll question explicitly asks about "positions", not "scrambles". I wonder how many of the other "No" voters made the same mistake.

Probably most, I don't think many people at all would want to include the solved position.

"People also pointed out that the Pyraminx Single WR was 6 moves which seems to be true while I thought it was 7 moves before. I agree that we should change Skewb and Pyraminx from 7 to 6.".

This really just makes me feel like not much thought has gone into this at all.

 

[KongShou]

The "feeling" part is where I *speculate* about what non-cubers would find easy rather than having measured it. In my opinion, if the majority of average-intelligence average-health non-cubers can solve the majority of N-moves-scrambles easily, then N-moves-scrambles aren't good enough. And that can be measured (not exhaustively, of course, but if you do enough tests, you can find the answer with enough certainty).

A non cuber could easily solve it. So what? Why is that not good enough? This make it not scrambled? Is this your definition of a scrambled state?
A two move scramble is still random, albeit being easy.

 

[Stefan]

I'm not trying to do that, I'm merely saying that I think what you're saying doesn't hold much ground.

You did rephrase it to "because it does". That's not ok. But saying it doesn't hold much ground is alright.

This really just makes me feel like not much thought has gone into this at all.

Maybe, though no amount of thinking will ever be able to prevent all mistakes.
(except maybe if you think of how to destroy the universe and all future and succeed in doing that)

 

[Methuselah96]

Relying on technicalities to get more people to vote "yes"? That's just low.




No, I don't care about whether the resulting scrambles feel easy or not - a sune plus AUF on the 2x2x2 feels easy, but that's already 8 moves, and no reasonable filtering method should remove that. k pretty much provides an upper limit on what proportion of positions we will exclude (so k=1/1000 has us exclude no more than 1/1000 of the positions for a puzzle) and thus guarantees the maximum possible effect our filtering has on the distribution. We should probably either choose a nice round number, or choose a number that gives us a desired amount of exclusion on one puzzle - as an example, if we feel like 4-move 2x2x2 scrambles should still be allowed but no easier scrambles should show up, we could choose a k that falls in this range, which is about .000104 to .000607. At least, once we have made one arbitrary choice, the rest of the limits follow naturally.

That was actually my intention when writing the choices for the poll. If I could redo the poll it would be:
Should the scrambling procedure include filtering certain positions?
Yes, the filters already in the regulations are perfect.
Yes, the positions filtered should be more than one move away for some puzzles, but not the positions filtered in the Regulations (changing Regulations 4b3b-4b3d).
Yes, the positions filtered should only be one move or less away (deleting Regulations 4b3b-4b3d).
Yes, the position filtered should only be the solved state requiring zero moves to solve (changing Regulations 4b3, 4b3e; deleting Regulations 4b3b-4b3d).
No.

Also I like your idea of a constant to make it more uniform.

 

[Kirjava]

You did rephrase it to "because it does". That's not ok.

When you're talking about how you /feel/ about it, that's essentially what it amounts to to me. I did not see this is misrepresentation, sorry if it felt so.

Maybe, though no amount of thinking will ever be able to prevent all mistakes.

Of course, but fact checking should be done for something as important as this.

 

[Stefan]

A non cuber could easily solve it. So what? Why is that not good enough? This make it not scrambled? Is this your definition of a scrambled state?

A random cube is random. So what? Why is that good enough? This make it scrambled? Is this your definition of a scrambled state?

Note that I didn't highlight Erik's "A scramble is no more than a random position of a cube in my opinion" like that just for fun. He himself understands and acknowledges that it's only his opinion (and apparently it's also yours). I have a different opinion. Explain why yours is valid and mine isn't.