Reply To: Scores by playing platform

Home Forums General Discussions Forum Scores by playing platform Reply To: Scores by playing platform

Kelani
@kelani

@captrec I also used to delay posting scores until a few ladies in the Bloated Pig told me posting sub-average scores prevents the average from becoming inflated, which is a problem for levels steamrolled by daily challenges, and those with fewer entered scores. They were right. I’ve entered a low score and seen the average drop by 20-100 points. If more people followed suit, the averages would drop quite a bit.

The iOS domination you mentioned is exactly why I need the actual platform distribution. I can compensate for it and find a true baseline, which will be a template for finding cursed levels. For example:

AB Space Overall 1st-2000th Place by Platform Percentage.
Any totals under 100% are due to no platform specified.

-----------------------------
Rank       iOS, And, PC
-----------------------------
1-100       75, 15, 10
101-200     64, 24, 12
201-300     71, 19, 9
301-400     78, 13, 7
401-500     56, 30, 13
501-600     66, 22, 10
601-700     66, 20, 13
701-800     61, 20, 12
801-900     70, 20, 9
901-1000    71, 17, 10
1001-1100   67, 16, 14
1101-1200   54, 28, 12
1201-1300   47, 34, 10
1301-1400   55, 31, 9
1401-1500   56, 19, 17
1501-1600   46, 28, 16
1601-1700   59, 24, 9
1701-1800   49, 25, 23
1801-1900   40, 24, 19
1901-2000   41, 21, 16
-------------------------

Top 2000 Average: 60% iOS, 23% Android, 13% PC, 4% Other
Top 100: 75% iOS, 15% Android, 10% PC
————————-

Using those ratios, I’d expect to find a similar ratio on any episode or level. If a level’s Top 100 is 35% iOS, 59% Android, 6% PC, IMO, that’s a red flag. As far as what % variance is statistically significant, I have no idea. A common threshold is 5%, but I don’t know if that’s appropriate here. I was just a music major, so for that, I must defer to any ABN math gurus who want to chime in.

The Top 100 ratio may be useful for quick-checking a level, but the Top 2000 ratio broken into 100-place chunks is better. At Episode/Game level, it can determine if any cursed levels exist. For a specific level, it does the same, in addition to determining if (and how) a platform’s scores may be affected. Take any level’s Overall Top 2000. If the expected breakdown is 75,15,10, but 1-100 is: 40% iOS, 37% Android, 23% PC, it suggests a quirk that handicaps iOS scores, or boosts scores for Android/PC.

Anyway, it’s a work in progress, and since I’m NOT a math guy, it’s highly likely my approach is flawed. I expect most cursed levels will be reported by talented flingers. As I have no way to test Android, I’d prefer to use math to confirm them. Also, it’d be much less work than manually running each candidate level through TestBox for a week. :)