Adaptive clan war matchmaker

Quote:

---

Originally Posted by dorsan

Problems:
- the matchmaker is manually adjusted with frequency low enough for players to exploit the system
- buildings weights do not accurately reflect their contribution to the war success of a clan
- offensive strength isn't weighted strongly enough to create fair wars

---

In general, the idea of adaptive weighting is decent conceptually, but would be very hard to make work in practice. But regarding the part I bolded, you have no way of knowing this since offensive "weight" isn't advertised anywhere (unlike defensive "weight," which people approximate based on the stealable loot in storages on battle day).

Even your 2nd "problem" is pure conjecture on your part, since NO ONE outside supercell knows the exact matchmaker weighting algorithm. The thing that contributes to the "war success of a clan" more than anything else in the world, hands down, is player skill, not "inaccurate building weights."

Quote:

---

Originally Posted by PhilISuppose

Eventually buildings (walls, heroes and troops too) will have an effective war weight that means something based on empirical data rather than the subjective opinions of the developers.

---

no what it means is that the preferred buildings of War farming clans will be severely OVER weighted (cause they keep losing wars), and things like inferno's and xbows would be severely UNDER weighted because the 1st thing bad TH10 rushers plop down is double infernos; they then go one to lose every war (cause they're unskilled rushers) and drag the weight of the very best defensive structures down.

You don't end up with accurate quantifiable weights for every building, you end up with building weights that ebb and flow based on the skill level, or lack thereof, of the average clasher (note, the average clasher is not very good). Do you really want some kid with a rushed TH10, double lvl1 infernos, gold walls, and who hasn't unlocked "golem" in his dark barracks contributing to the quantifiable algorithm that determines clan war matchups? I certainly don't.

Quote:

---

I saw a list once that gave weightings for heroes on a purely linear scale. It made no note of the huge difference every five levels of hero makes. If that's truly how heroes are weighted, then they need all the help they can get.

---

As a general rule, I wouldn't trust everything I see on the internet. I have seen plenty of spreadsheets and posts from people claiming to have the holy grail of "war weight" calculations, and 99.9% of them are very wrong. Moskri's spreadsheet (or whatever that guy's name was) was passed around as gospel for months, we STILL see people referencing it on the forums, but it was VERY wrong.
No one outside of supercell knows the exact way "war weight" and clanwar matchmaking is done, and I think they prefer it this way.

Quote:

---

Originally Posted by nerfedname

In general, the idea of adaptive weighting is decent conceptually, but would be very hard to make work in practice. But regarding the part I bolded, you have no way of knowing this since offensive "weight" isn't advertised anywhere (unlike defensive "weight," which people approximate based on the stealable loot in storages on battle day).

Even your 2nd "problem" is pure conjecture on your part, since NO ONE outside supercell knows the exact matchmaker weighting algorithm. The thing that contributes to the "war success of a clan" more than anything else in the world, hands down, is player skill, not "inaccurate building weights."

---

The matchmaker is not voodoo magic. It is in fact simple enough to reverse engineer it based on a very limited data from less than 100 wars. The only thing that makes it harder to get the exact weight values is the fact that the weight is rounded before displayed to us. And there were plenty of experiments with engineered clans that clearly showed that your offensive strength has little to no effect on the matchmaker.

Quote:

---

Originally Posted by dorsan

The matchmaker is not voodoo magic. It is in fact simple enough to reverse engineer it based on a very limited data from less than 100 wars.

---

then why, in the 18 months that clan wars have been available in game, has no one been successful at accurately reverse engineering this calculation?
If it was so easy, there should be a widely circulated, completely accurate spreadsheet at this point.

Quote:

---

The only thing that makes it harder to get the exact weight values is the fact that the weight is rounded before displayed to us.

---

well that, as well as the fact that offensive weight is not displayed or advertised anywhere, at all.

Quote:

---

And there were plenty of experiments with engineered clans that clearly showed that your offensive strength has little to no effect on the matchmaker.

---

Then why does Supercell say that offensive upgrades (e.g., camp space, spell factory, hero levels, troop levels) contribute to matchmaking?

I have seen forum posts by those claiming (always anecdotally) that THX.5 is the way to go. This usually involves some discussion of how easily they can 3star their "mirror" or where they fall on the war map, both of which are entirely irrelevant and prove nothing.

I have also seen forums posts by those claiming (always anecdotally) that upgrading their spell factory (to unlock freeze, for example, in a TH9.5) caused them to start pulling harder "mirrors" and therefore upgrading the spell factory contributes a lot to one's matchmaking weight (as I would expect it to, however anecdotal evidence proves nothing).

Only supercell knows how war matchmaking is done, and from the players perspective it is a complete black box (we know next to NOTHING of how it's conducted). Supercell specifically states that offense contributes, you're going to have to take their word on that. And they have the metrics on clan war matchups that no one else has access to; when they observed that hero levels and wall levels were undervalued in their algorithm they changed the relative contribution of those two items (update last April I believe). If at a later date they have the metrics to prove, as many unsubstantially claim, that offense is undervalued they will change that as well.

Edited to add: And again, I do like an "adaptive matchmaking system," at least conceptually, I just think it would be VERY hard to implement properly because far and away, the #1 thing that contributes to the success of a clan in war is NOT "how many offensive upgrades they have," or "how many THX.5's they have," or "how many bases have infernos." The #1 thing that contributes to a clan's success in war is player skill. Supercell cannot quantify that in an algorithm.

Quote:

---

Originally Posted by nerfedname

no what it means is that the preferred buildings of War farming clans will be severely OVER weighted (cause they keep losing wars), and things like inferno's and xbows would be severely UNDER weighted because the 1st thing bad TH10 rushers plop down is double infernos; they then go one to lose every war (cause they're unskilled rushers) and drag the weight of the very best defensive structures down.

You don't end up with accurate quantifiable weights for every building, you end up with building weights that ebb and flow based on the skill level, or lack thereof, of the average clasher (note, the average clasher is not very good). Do you really want some kid with a rushed TH10, double lvl1 infernos, gold walls, and who hasn't unlocked "golem" in his dark barracks contributing to the quantifiable algorithm that determines clan war matchups? I certainly don't.

---

You are making 2 mistakes here.

1) You are assuming that all bad players will follow the same upgrade patterns
and
2) You are assuming that as soon as infernos won't become underweighted pro clans won't start mass building them and then dragging down their weight as a result

The beauty of an adaptive system is that whatever the situation is, it will do just what it's name says.. Adapt. :)

Quote:

---

Originally Posted by dorsan

The beauty of an adaptive system is that whatever the situation is, it will do just what it's name says.. Adapt. :)

---

I do understand what you're going for, but the system adapting is not always preferable.

Again, far and away, the most widely fluctuating variable in clan war success is PLAYER SKILL, not "how much a lvl1 inferno weighs relative to a lvl 2 xbow". A system that adapts to a widely fluctuating variable, that cannot be calculated, quantified, or controlled, is not necessarily a good thing.

Edited to add: Now I'm having flashbacks to undergraduate "process systems controls" class.
:shudder:
:)

Quote:

---

Originally Posted by nerfedname

...too much to quote at once...

---

I have recently started gathering information about the actual offensive strength for enemy clans to see how much it affects the matchmaker.

These are the premises under which I gather my data:
1) we can calculate the offensive strength of clans by comparing loot on battle day
2) there is a fixed limit for mismatch between clans if the matchmaker didn't take longer than 5 minutes to find a match

If the offense is weighted the same as defense, matches with huge offensive mismatch will either have huge defensive mismatch to the other side or no mismatch. If both offense and defense are weighted the same or near same it would be impossible for the matchmaker to find matches where both offense and defense are mismatched to the same side.

However, this isn't what I am observing. The match we had the hugest defensive mismatch (12%) also had the largest offensive mismatch to the same side as well. Only way to justify this is to say that the mismatch threshold is larger than 12%, but in that case, sooner or later a 20-25% defensive mismatch should appear with no offensive mismatch. If it does not, we can conclude that the offensive weight is disproportional to the defensive weight.

Quote:

---

Originally Posted by nerfedname

I do understand what you're going for, but the system adapting is not always preferable.

Again, far and away, the most widely fluctuating variable in clan war success is PLAYER SKILL, not "how much a lvl1 inferno weighs relative to a lvl 2 xbow". A system that adapts to a widely fluctuating variable, that cannot be calculated, quantified, or controlled, is not necessarily a good thing.

---

However, while player skill plays role in individual war wins or losses, on a large sample it will only generate noise, unless all the successful clans follow a preferred upgrade pattern - which then rightfully becomes the upgrade pattern with the highest weight.

Quote:

---

Originally Posted by nerfedname

no what it means is that the preferred buildings of War farming clans will be severely OVER weighted (cause they keep losing wars), and things like inferno's and xbows would be severely UNDER weighted because the 1st thing bad TH10 rushers plop down is double infernos; they then go one to lose every war (cause they're unskilled rushers) and drag the weight of the very best defensive structures down.

---

I think you vastly over estimate the number of War Farming clans out there.