The one and only " War search after update" thread

[Vikingchief]

If we go off of what Darian said, we now have two groups. And that is the only possible calculation the MM can do. The MM is basically a calculator, it's adds up numbers, these numbers, add up to one number for the clan, it then matches clans that are within those parameters. We call them weights, to the MM it's just a value.

It now has a ratio of defense vs offense to separate lopsided defenses.

The only "limiter" the MM has, is the total weight of the clan. If it does not find a match within a creation amount of time, it then increases or decrease the value (weight) of the clan so as to match it with another.

There's nothing else the MM does.

you dont know this, you are claiming it does, If you want to have a decent discussion you cant just assert things, you have to explain how you know what you know, what evidence exists to prove what you are saying.

[blombardo]

Ok..
10x engineered war -- 100% fair (tie/draw)
10x engineered war #2 -- 95% fair (tie /draw)
30x engineered war - not quite fair... enemy top bases had 5 max tens our top bases had th8/9 ish defenses. But we beat them on heroes big time. -- Stats for the top 8: This war ended as a tie 1 star short of perfect with a .03 percent difference in our destruction... of course -- they had to clear a 79k base... WE had to clear an 89k with infernos... big difference in what we had to hit...

Draconia	11	79	45	45	20	#22PJJ8PVU	110	-10	51	1	aliostovan	11	89	22	27	10	max laloon for th11 /engineered level 9 and 10 walls
Bleazel groupme = band	11	76	45	45	20	#29QGY2CLV	110	-12	41	2	amirhosein	10	88	29	40	0	max ten / max troops and walls for ten
Happy2	11	70	45	45	18		108	-16	63	3	cant read it	10	86	18	27	0	85% max troops for ten and max walls
Ebrehnr	11	62	34	45	20		99	-18	77	4	mohammad73	10	80	11	11	0	skull walls th9 troop levels rushed ten with max air def
Raven	11	62	17	17	11	#2Y8VCG2CR	45	-8	10	5	saeed	10	70	17	18	0	th9 troops and level 9 walls (*max 9 normal)
RaginTalon	9	58	10	10	0		20	-11	-16	6	hojat	10	69	20	16	0	max 9 research all level 10 walls
Nightmare 10 weight-9-15	11	57	20	20	10		50	-4	35	7	cant read it	9	61	8	7	0	level 4 drags only /rushed pink walls
Nightmare 09 (th9 mirrors now)	11	57	20	20	12		52	-4	27	8	cant read it	10	61	12	13	0	level 5 dragon zap only - th6 walls pink missinglots

TODAYS war
10x - 95% fair engineered
30x -normal war - 90% fair
30x engineered war - INSANELY mis matched - worse than any war I have ever seen in my 4 years of clashing ; an dI've seen a lot of bad matches ... this ones insane...
Enemy top 10 are heavily engineered max 10.5's with 100% max offense (level 15 double archer/cannon/ inferno etc.. no eagles)
our top 2 are heavily engineered at a whopping 79k with no inferno or eagle.. (easy fodder for any one of those bases)
our next few have mid th10 strength and th8/9 weight..

can you say boned? ♥♥♥♥... This is after I removed all my mini accounts out and used only 2 of my own... we're up against a 35 win streak clan - and I see why ... they have a super solid top.. and guess what?

they have 15 fluffy th3 bases on bottom with 1 cannon no walls and no troops...

WE have zero accounts like that.

--So there you go everyone... if you wanna engineer a war win streak like that -- grab a pile of maxers -- and throw in a pile of useless th3's on bottom and BOOM... you get an epic advantage...

Time for me to make 20 new town hall 3 accounts I guess...

then the enemy will win if they have 25...

so maybe I'll just hve to make 47 town hall 3 accounts then put 3 maxers on top... Then perhaps I'll have a shot at winning...? ♥♥♥♥

this whole mess is supercells fault and I hope they fix it soon...

[SuperStorm103]

Heh, just as I and many other feared. We saw this potentially coming from a mile away. Apparently, SC is not aware of roster engineering. And we all thought there was no way they could be that out of touch. Gotta love averages!

/facepalm

[Shaloob]

What the hell update does supercell made in warmatchmaking...we started war even without any engineered or .5s . I feel really irritated on seeing an engineered th11 with maxed loons in enemy. Easy to hit us as we are just th9s in top.

[posthoc]

There isn't a single system that can be improved by taking a continuous variable and turning it into a binary yes/no (in this case, engineered or not). The MODELS of those systems might improve, but the systems won't. So I call ♥♥ on your theory iBork,

[Mico11813]

2nd war since the update, both matched in under 2 mins and both have been fair. I guess we're getting lucky.

Running 10v10, a couple of mid/max 10s, mid/maxed 9s, and a couple of high/mid level 8s and we're matching very similar clans.

[AntB77]

Our 2nd war for us and it's just like the first and we will win. As a .5 clan we're facing heavily engineered th11's (8+ in a 20/20 war) with low def and max war attacks in both wars. Never had these type of opponents before, usually get lighter engineering or .5's.

TH11's have 2 attacks, spam max miners or a really bad spam loons lavaloon. They smash everything up to a low TH10, then fail miserably higher. No skill, all they can do is rely on mismatches and judging by their win streaks we've broken, they had that advantage up until now.

[Sservis]

If we go off of what Darian said, we now have two groups. And that is the only possible calculation the MM can do. The MM is basically a calculator, it's adds up numbers, these numbers, add up to one number for the clan, it then matches clans that are within those parameters. We call them weights, to the MM it's just a value.

It now has a ratio of defense vs offense to separate lopsided defenses.

The only "limiter" the MM has, is the total weight of the clan. If it does not find a match within a creation amount of time, it then increases or decrease the value (weight) of the clan so as to match it with another.

There's nothing else the MM does.

This is something that would be very easy for cocp.it to check. Use the graphs defined by nodes of clans and edges of war matchups. Does the graph show a more distinct separation into groups than before. Sounds like a good blog post for them. The hard part is figuring out how to define regions of the graph, but I'm sure with a quick google, there exists research devoted to measuring the number of semidistinct regions in a graph/network.

[Sservis]

There isn't a single system that can be improved by taking a continuous variable and turning it into a binary yes/no (in this case, engineered or not). The MODELS of those systems might improve, but the systems won't. So I call ♥♥ on your theory iBork,

Yes, the accuracy is never improved by throwing away lower order details. Other factors may be, calculation difficulty for instance. Or search time. offense/defense ratings for each base is many more comparisons to "get right" than clan totals, so the threshold has to be relatively wider or it will take longer ie more even overall [but potentially off a lot at a few bases] or more even at each base [but off more in total]?

Personally, it seems like they didn't tweak much. They definitely increased search times, which suggests they're compromising less on match quality for a given amount of searching, maybe also with a lower cap on "worst acceptable match" This could be potentially be enough to explain nearly all the results without any specific tweaks to the formula. However, if they did this, I'd have expected them to decrease prep day also. Not much difference between 23 hours and 20 hours.

Edit: I think it likely that they also made other tweaks, like increasing offenses importance in a match being considered "acceptably close"

Edit2: I'm not saying iBork is right (I don't know), but the reasons you pointed out that iBork was wrong aren't sufficient. I do agree with you that separating clans into two bins seems like a very poor solution that introduces a new form of error (how bases/clans are classified) with a discontinuity in the output function of matchmaking when clans transition between the two sets. Discontinuous functions are very often bad game design.

[shkata]

There isn't a single system that can be improved by taking a continuous variable and turning it into a binary yes/no (in this case, engineered or not). The MODELS of those systems might improve, but the systems won't. So I call ♥♥ on your theory iBork,

There is no need for complex models and systems..
Just give some value for the offense and defense, count the best 2 attacks against each base. Then when you have 2 attacks and 2 defenses plus defense and offense weight ... and everything will go back to normal. Yeah ... the perfect wars will be very rare... but people will start building defenses , the bases will be more similar, the matches will be more fair , roasters will be more similar, war search time will decrease...