while i appreciate all the effort spent on this analysis, you guys are barking up the wrong tree. real cw issue is not tie breaking but reward structure.

rewards could and should be so easy:
1. total # of territories in season map = 100%
2. total prize pool = 10k coins
3. coins alloted to clans according to % of territories won, rounded to full number, 0.5 = rounded up
4. distri of coins in clan alloted to players according to their % of clan total wins (free and regular wins)
5. discretionary/idle awards: tier 1 for every clan with 20% or more tof tital territories; tier 2: 10-19%, tier 3: 5-9%, tier 4: all others: 1 uncommon arti, forget about the ridiculous free cache

257 beat 513 (both with a mod value of 1)

f fizzer!

Edited 1/25/2023 19:52:52

my reward model would level the playing field, give smaller clans their deserved share of the spoils while still rewarding "elite" clans

ah yes,and awarded artifacts should be "of player's choice".

haha downvote my posts as much as you like, little farters

while i appreciate all the effort spent on this analysis, you guys are barking up the wrong tree. real cw issue is not tie breaking but reward structure.

rewards could and should be so easy:
1. total # of territories in season map = 100%
2. total prize pool = 10k coins
3. coins alloted to clans according to % of territories won, rounded to full number, 0.5 = rounded up
4. distri of coins in clan alloted to players according to their % of clan total wins (free and regular wins)
5. discretionary/idle awards: tier 1 for every clan with 20% or more tof tital territories; tier 2: 10-19%, tier 3: 5-9%, tier 4: all others: 1 uncommon arti, forget about the ridiculous free cache
6. all artis "of players' own choice"

Edited 1/25/2023 20:10:22

rewards could and should be so easy:
1. total # of territories in season map = 100%
2. total prize pool = 20k coins
3. coins alloted to clans according to % of territories won, rounded to full number, 0.5 = rounded up
4. distri of coins in clan alloted to players according to their % of clan total wins (free and regular wins)
5. idle awards canceled, as Idle is cancer.

Edited 1/25/2023 21:26:31

If you write in a document all (or sample bug enough) tiebreakers and their respective clan IDs, I will make a program that will test your mod 128 hypothesis.

Edited 1/25/2023 21:29:36

@NONO, That tree has been barked up as well

Yeah. Is there another tiebreak?

Yes, of course there is - and it is simply the order in which the data exists in the source tables. If all other sorting mechanisms don't provide a distinctive order, it will just be whatever the order of the data in the primary table joins on to correlate data from the others ... so you need to know the query & table structure in order to identify the order.

Better to just request something that makes sense.

But I'm down with revamping the reward structure to.

HOWEVER we also know that Fizz thinks rewards should be common to the clan, not individual. I think minor individual rewards should be used to generate motivation, then have the larger rewards for the whole clan. Needs to be something to motivate some players to do more than just 1 game per season and ride on the laurels of the active CW roster. So need to be more convincing on this point to make a change there.

By the way, you just need a random sample of 5 with all negatives to confirm, that "ID mod 128" affects tiebreakers (for 0.95 confidence).

Edited 1/25/2023 21:32:45

My google sheets link had the data, but here is another different link only with it:
https://pastebin.com/VsiUNtJF

I'm curious if mod 128 has stronger likelihood for all clans that existed prior to clan wars. I find a good amount of the exceptions with the higher clan IDs, say >550. maybe see how the odds change with full data set vs excluding clan IDs >550

Edited 1/25/2023 21:42:06

(306, 411, [(3, True), (11, False), (12, True), (14, False), (15, True), (21, False)])

Could you explain, please?

I referenced that one in my response to knyte:

"A curious case is between clans 306 and 411, where they tied 6 times, but the results alternated. I will take this as the exception that proves the rule lol. The around 90% prevalence of consistent tiebreaker order does suggest this being a somewhat predictable trend compared to a completely different random ranking every season."

so this means that season 3 306 won tiebreaker, but season 11 lost?

Yes that is correct, also from my response to knyte on page 1...

"You'll see cells filled with info like: (41, 197, [(16, True), (17, True), (19, True), (20, True), (22, True)])
where the first two numbers are the 2 clan ID, then whether the first clan ID won the tiebreaker in given seasons."

(507, 251, [(18, True), (20, True), (22, True)])

(259, 515, [(1, False), (16, False), (17, True)])

(41, 297, [(20, True), (22, False)])

(259, 387, [(5, True), (12, True)])

(306, 562, [(14, True), (15, False)])

(261, 517, [(14, True), (24, False)])

(52, 436, [(2, True), (3, True)])

(251, 635, [(15, False), (23, False)])

(11, 267, [(14, False)])

(137, 9, [(18, True)])

(257, 513, [(9, True)])

(20, 276, [(11, False)])

(170, 682, [(22, True)])

(323, 67, [(13, False)])

(511, 255, [(17, True)])

(41, 681, [(17, True)])

(197, 325, [(3, False)])

(26, 282, [(11, True)])

(94, 350, [(16, True)])

(188, 572, [(20, True)])

(261, 389, [(14, False)])

(383, 255, [(2, True)])

(515, 3, [(6, True)])

(516, 388, [(2, True)])

(155, 411, [(3, True)])

(84, 596, [(12, True)])

(455, 199, [(1, True)])

(51, 691, [(22, True)])

(517, 389, [(14, False)])

Lots of matches with mod 128.

@re: "Why mod 128 (2^7) and not mod 256 (2^8)?"

it may be due to how Postgres encodes variable-length integers. Some common encoding schemes for large integers have 7 bits of data and a continuation bit to indicate whether or not to look at the next byte: https://en.wikipedia.org/wiki/LEB128

I don't know if Postgres uses LEB128 in particular, but if it uses something similar that could explain the ordering. What's interesting now is how CW would treat pairings with the same mod-128 value, and of course the mystery of 306/411.

I will expect that the tiebreaker has an alternative distribution with parameter p. True = clans with lower mod 128 value has won, False clan with bigger mod 128 value has won. Results are independent of each other. I will ignore tiebreakers with the same mod 128 value. If p =0.5, mod 128 has no impact. If p = 1, you always win, if p=0, you always lose. High values of p mean that having a lower mod 128 value helps a lot.

There are 6037 samples and the result of True can be found 5275 times. 0.95 confidence (= "standard" in statistics) interval for p: (0.8651626 0.8819187), where the most likely value of p is 0.87.

In other words, having a lower mod 128 ID helps A LOT.

Why mod 128 (2^7) and not mod 256 (2^8)?

The best p can be achieved for 256: p = 0.92 ± 0.007. So yep, this is a VERY good question.

Edited 1/26/2023 15:01:16