But how do you explain that I’m able to use 2x 4* 5/50 and 1x 5* highest rank to lowest rank with no death matches? Because the limit for me is 15k and I know if I go below 15k then I face death matches. The amount of 5* I have from highest rank to lowest rank with mix match combination using 2x 4* 5/50 allows me to use all those unused lower ranked 5* without having any problem and all of them is staying above 5*. Because it’s those 4* that kinda boost it up to remain above 15k threshold and the cooldown is 2 hours for those 4*. Using the same method with my 6* collection.
15000 is within the margin for error of the 14600 PI below which the analysis predicts will generate death matches.
I don’t know about you, but I’ve used this method for about a year now. I have no idea what you’re talking about.
I'm not sure what your issue is. You said "because the limit for me is 15k and I know if I go below 15k then I face death matches." That sounds about right to me: my analysis predicts the same thing. As I mentioned in the article, while I give calculated numbers, those numbers do have some margin for error within them inherent in the calculations themselves, beyond the systematic sources for error that might also exist and still need to be studied.
But the fact that your infinite streak system seems to be based around the notion that going below 15k is dangerous is consistent with my analysis that says going below 14550 is dangerous, within the fuzzy limits of precision I can currently claim.
Second: I still haven't done what I would say is a complete analysis of how masteries affect every aspect of the arena system. They seem to affect different parts differently, and I do not know if masteries affect how the game determines if a match can be located. Part of the reason for releasing this article was my home that people willing to test the theory under a wide range of different roster and mastery conditions might offer some better insight into how those variables impact the system as a whole. It would take prohibitively long to test every possible combination of factors single handedly.
I want to remove all masteries and run the bottom of my roster as I normally would to prove/disprove our assertions relating to PI. However that will not be done for some time as I can’t afford to drop the units for a test but will do so if I can remember when I drop suicides again. Will report back if done but at a much later time.
*Actually I’m running suicides now but haven’t run them regularly for quite some time and even prior I never hit a death match running the bottom of my roster based on my naked pi. The bottom of my roster with suicides is 15-16.6k which drops significantly without suicides and I want to say well below the 14.6k.
I can’t say the prevalence of doom in death matches means anything more than he is the most common high base pi rank up. Gambit and silver surfer are, imo, far less likely to make it to rank 3. Additionally when teams are displayed they display in order of descending PI not naked PI and doom is at the top of every pi list aside from naked pi.
It is possible that if a champ doesn't show up, it is because the arena is somehow "unaware of it." But there's a way to account for this possibility as well: if Gambit shows up *somewhere* but then doesn't show up within the top tier Death Matches, that would imply the game believes Gambit is "good enough" to send at players but not "good enough" to send at players asking for the highest possible matches, which would imply within the criteria being used by the arena he's lower than other options.
For this to be valid, Gambit has to show up as a competitive max rank max sig champ, meaning 5/50 sig 99 in the 4* arena or R3 sig 200 in the 6* arenas. The former might be easier to catch in the wild than the latter. This is not something I specifically looked out for, but I will be specifically looking for it now.
I can’t say the prevalence of doom in death matches means anything more than he is the most common high base pi rank up. Gambit and silver surfer are, imo, far less likely to make it to rank 3. Additionally when teams are displayed they display in order of descending PI not naked PI and doom is at the top of every pi list aside from naked pi.
It is possible that if a champ doesn't show up, it is because the arena is somehow "unaware of it." But there's a way to account for this possibility as well: if Gambit shows up *somewhere* but then doesn't show up within the top tier Death Matches, that would imply the game believes Gambit is "good enough" to send at players but not "good enough" to send at players asking for the highest possible matches, which would imply within the criteria being used by the arena he's lower than other options.
For this to be valid, Gambit has to show up as a competitive max rank max sig champ, meaning 5/50 sig 99 in the 4* arena or R3 sig 200 in the 6* arenas. The former might be easier to catch in the wild than the latter. This is not something I specifically looked out for, but I will be specifically looking for it now.
There are quite a few factors in the strengths of teams since they are teams.
If doom commonly makes it to teams with other high naked pi champs and gambit makes it to teams with lower naked pi it skews what will be returned for matches.
Second: I still haven't done what I would say is a complete analysis of how masteries affect every aspect of the arena system. They seem to affect different parts differently, and I do not know if masteries affect how the game determines if a match can be located. Part of the reason for releasing this article was my home that people willing to test the theory under a wide range of different roster and mastery conditions might offer some better insight into how those variables impact the system as a whole. It would take prohibitively long to test every possible combination of factors single handedly.
I want to remove all masteries and run the bottom of my roster as I normally would to prove/disprove our assertions relating to PI. However that will not be done for some time as I can’t afford to drop the units for a test but will do so if I can remember when I drop suicides again. Will report back if done but at a much later time.
*Actually I’m running suicides now but haven’t run them regularly for quite some time and even prior I never hit a death match running the bottom of my roster based on my naked pi. The bottom of my roster with suicides is 15-16.6k which drops significantly without suicides and I want to say well below the 14.6k.
I tried once, only got to round 11. Playing without any masteries at all is like practically playing a completely different game, and I discovered even Parry significantly inflated PI. I'm not even sure I can get to round 20 with no masteries at all consistently, but this is part of a long term project.
But how do you explain that I’m able to use 2x 4* 5/50 and 1x 5* highest rank to lowest rank with no death matches? Because the limit for me is 15k and I know if I go below 15k then I face death matches. The amount of 5* I have from highest rank to lowest rank with mix match combination using 2x 4* 5/50 allows me to use all those unused lower ranked 5* without having any problem and all of them is staying above 5*. Because it’s those 4* that kinda boost it up to remain above 15k threshold and the cooldown is 2 hours for those 4*. Using the same method with my 6* collection.
15000 is within the margin for error of the 14600 PI below which the analysis predicts will generate death matches.
This is true. I just tested it for me self. I think I didn’t give it an extra thought because the highest r5/50 4* kept refreshing allowing me to mix match with my 5* and 6* collection of all ranked stats. I often land from 17-24k depending what the rank of the 5* and 6* is.
That's some effort to do this write up. My query for a long time has been why kabam can't just program so that all the match ups are Kang and thanos teams having similar pi to the team that we have chosen. Get rid of the death matches and allow us to use our entire roster in the arena. It also has the advantage of not ranking up 5 stars to r3 specifically for arena.
what am I missing? A 16798 team returned a death match?
Probably the rank stats. Would be different if your 4* was r5. Even if the r4 exceeded beyond the safe spot still somehow it trigger deathsquad so the ranking might have some connection to it with the PI cooking pot to avoid death matches.
what am I missing? A 16798 team returned a death match?
This may be the result of those random factors I mentioned. This is not something I'm certain about, but I believe that the game always fudges its "target" that it is looking for by a small random amount to "shake up" the matches a bit. This *might* mean that when the game should be looking for a 3.9x match and ordinarily wouldn't find one, it might be looking for a 3.7x or a 4.1x in any one particular round. That random variation might make death matches pop up sometimes higher and sometimes lower than some "target" value.
Another possibility: since different players do not tend to get the exact same matches, but one player grinding a lot can often notice seeing the same guys over and over again, this suggests that the matches I get and the matches you get aren't coming from some gigantic random pool. Instead we are each seeing some smaller, maybe hash table subset of all possibilities that could be different between us (that may also explain why in half a million fights of arena grinding I've never matched against myself). So it is possible that there is some small variations in what is possible for me and what is possible for you, and it is also possible this may change over time. So what's "safe" and "not safe" might be blurred by these additional dynamic factors.
This may be something that will need to be characterized over a lot of testing.
what am I missing? A 16798 team returned a death match?
Probably the rank stats. Would be different if your 4* was r5. Even if the r4 exceeded beyond the safe spot still somehow it trigger deathsquad so the ranking might have some connection to it with the PI cooking pot to avoid death matches.
Yet this does not return a death match.
The difference between the two? The first was less than 25% of the max naked pi available in arena (gambit, doom, surfer) and the second was greater than 25% of the max naked pi.
what am I missing? A 16798 team returned a death match?
This may be the result of those random factors I mentioned. This is not something I'm certain about, but I believe that the game always fudges its "target" that it is looking for by a small random amount to "shake up" the matches a bit. This *might* mean that when the game should be looking for a 3.9x match and ordinarily wouldn't find one, it might be looking for a 3.7x or a 4.1x in any one particular round. That random variation might make death matches pop up sometimes higher and sometimes lower than some "target" value.
Another possibility: since different players do not tend to get the exact same matches, but one player grinding a lot can often notice seeing the same guys over and over again, this suggests that the matches I get and the matches you get aren't coming from some gigantic random pool. Instead we are each seeing some smaller, maybe hash table subset of all possibilities that could be different between us (that may also explain why in half a million fights of arena grinding I've never matched against myself). So it is possible that there is some small variations in what is possible for me and what is possible for you, and it is also possible this may change over time. So what's "safe" and "not safe" might be blurred by these additional dynamic factors.
This may be something that will need to be characterized over a lot of testing.
I think anyone who’s been around a while can attest to there being different tables or pools of players which you can only pull from. This can cause differences in max potential for sure. It’s also why we repeatedly face the same players, we pull from a limited set list rather than an unlimited list.
what am I missing? A 16798 team returned a death match?
Probably the rank stats. Would be different if your 4* was r5. Even if the r4 exceeded beyond the safe spot still somehow it trigger deathsquad so the ranking might have some connection to it with the PI cooking pot to avoid death matches.
Yet this does not return a death match.
The difference between the two? The first was less than 25% of the max naked pi available in arena (gambit, doom, surfer) and the second was greater than 25% of the max naked pi.
I'm not sure I understand "naked PI" correctly. You said this is the PI of the champ unawakened, with no masteries or other boosts, at the rarity and rank in question. So the max team available would be 6* R3 unawakened Gambit, Doom, and Silver Surfer. According to Auntm.ai, that would be 11289 + 10434 + 10298 = 32021. So that's the max PI available in the arena according to your theory.
When I do the math for the team that did, and did not trigger deathmatch I get 2966 (Falcon 4/40) + 2699 (Gamora 4/40) + 2862 (Ultron 4/40) = 8527 as the naked PI for the team that did not trigger death match and 2957 (DDHK 4/40) + 2915 (Sasq 4/40) + 2914 (Mr F 4/40) = 8786. Both teams appear to have higher naked PI than 25% of the maximum naked PI (which would be 8005).
Also, the naked PI of the death match team you did pull is (if I understand naked PI correctly) 30152. That's about 3.54x the naked PI of the team you used. That seems odd, if I understand your naked PI theory correctly this team should have been closer to 4x your own team's naked PI.
What am I doing wrong with regard to computing naked PI value?
what am I missing? A 16798 team returned a death match?
Probably the rank stats. Would be different if your 4* was r5. Even if the r4 exceeded beyond the safe spot still somehow it trigger deathsquad so the ranking might have some connection to it with the PI cooking pot to avoid death matches.
Yet this does not return a death match.
The difference between the two? The first was less than 25% of the max naked pi available in arena (gambit, doom, surfer) and the second was greater than 25% of the max naked pi.
I'm not sure I understand "naked PI" correctly. You said this is the PI of the champ unawakened, with no masteries or other boosts, at the rarity and rank in question. So the max team available would be 6* R3 unawakened Gambit, Doom, and Silver Surfer. According to Auntm.ai, that would be 11289 + 10434 + 10298 = 32021. So that's the max PI available in the arena according to your theory.
When I do the math for the team that did, and did not trigger deathmatch I get 2966 (Falcon 4/40) + 2699 (Gamora 4/40) + 2862 (Ultron 4/40) = 8527 as the naked PI for the team that did not trigger death match and 2957 (DDHK 4/40) + 2915 (Sasq 4/40) + 2914 (Mr F 4/40) = 8786. Both teams appear to have higher naked PI than 25% of the maximum naked PI (which would be 8005).
Also, the naked PI of the death match team you did pull is (if I understand naked PI correctly) 30152. That's about 3.54x the naked PI of the team you used. That seems odd, if I understand your naked PI theory correctly this team should have been closer to 4x your own team's naked PI.
What am I doing wrong with regard to computing naked PI value?
You understand my naked pi thing correctly and You’re right about the math (cept ultron is 2682) I had overlooked falcon and ultron’s rework so I was operating on them being bottom of the barrel PI as they had been in the past.
Maybe the modifier floats as you suggested before. Or there is a slight rank/rarity modifier in addition to the 4x. However eliminating low naked PI from my r2 5s prevented me from ever encountering a death match while being able to mindlessly throw those champions into the arena. The 5* team of 13760 I showed has been routinely ran through arenas with zero death matches, even without suicide masteries which give it an even lower PI rating.
*the team in that picture isn’t the highest available team pi, it’s the lowest by about 11k, my options were NF or Elsa led teams (I have no idea who was paired with those teams, more on this in the spoiler) and doom was the easier match up so I choose him. The game obviously doesn’t solely choose only 4x teams but rather pulls from a variety of ~4x teams.
There are two things I’ve learned from arenas In relation to death matches. Minimum x PI doesn’t exclude you returning a death match, and ranking from the lowest naked PI will eventually eliminate the possibility of returning a death match when it comes to 5*.
Well maybe 3 things because the lower you go with naked pi the higher the chance of finding a death match, also the inverse, but that’s related to ranking the lowest naked pi. This was born of observing and analyzing countless death matches where the only common denominator was naked PI irregardless of displayed PI; pairing 2 low naked PI champs with the same rank/level could return a death match and only pairing 1 would not return a death match, additionally pairing 3 higher naked PI and low displayed PI wouldn’t return a death match. This to me eliminated displayed PI as a factor.
I ran the same falcon team again and returned two of the previous same teams. I choose the elsa team this time so show what was behind it. Naked pi was elsa-10002, domino-9853, doom-10434
I will run the same team when they come off cooldown and record the numbers and possibly sig level this time for true prestige. I also fought 6 other matches testing the plain 4* naked PI and returned two death matches contrary to my straight naked PI theory. Being this works as 5* and not as 4* at the same challenger rating suggests there might be and additional value at play.
Most of us have developed our intuitions of how death matches work in a similar way: eliminate champs that seem to not work, until we’re left with a set that seems to always work. However, it has left us all with different ideas of why they work, even if they consistently work for us. I’m hoping between those confirming my observations and those trying to disprove them, we might reach a more universal theory that explains everyone’s observations.
Maybe the modifier floats as you suggested before. Or there is a slight rank/rarity modifier in addition to the 4x. However eliminating low naked PI from my r2 5s prevented me from ever encountering a death match while being able to mindlessly throw those champions into the arena. The 5* team of 13760 I showed has been routinely ran through arenas with zero death matches, even without suicide masteries which give it an even lower PI rating.
It occurs to me there's a way to directly test the naked PI theory. If the theory is correct, then a team in which all three of the champions are unawakened should draw, more or less, the same matches that the same identical team at max sig would draw, since naked PI is unaffected by sig level. So if we were to form a team of unawakened champs and run them repeatedly in the arena, and then compare that to an identical team at max sig at the same round counters, we should see identical, or at least similar match ups. Conversely, if the match ups are radically different, that would tend to imply that sig level does matter to match ups, and is more likely to also play a role in triggering death matches.
We can't sig up and down champions, but I can in theory perform an analogous test by comparing two different accounts with champions at max sig vs unawakened. I'm not sure if I will have the time to perform this test this weekend as I might be tied up, but I'll throw this out there if this is something you want to tackle yourself (or anyone else wants to do so).
So, basically what Miike said last year, except much more detailed?
Some of this stuff has been floating around out there in different forms, but no one to my knowledge has actually articulated the specifics. To my knowledge, no one actually *knows* all of the specifics.
So, I managed to perform a test of the NakedPI theory. The best way I could think of to test this theory was to test the same team in the arena at max sig and unawakened and compare the results. I did this by finding a team I could run on two different accounts where the champions were at the same rank, but were unawakened on one account and max sig on the other. Furthermore, I did the test by running *only* that team in an arena, which is possible if you continuously refresh it. Slightly painful use of units, but I think worth it in this case, as I'm of the opinion that if @CoatHang3r believes it to be true, the odds of something being in there are sufficiently high to see for myself.
And as it turns out, that test proved worth the units:
This is the results from my main account, using the team of 4* Gwenpool 5/50 sig 99, Morningstar 4/40 sig 99, and Immortal Ironfist 4/40 sig 99. The three curves are the PI I faced in each arena round. And this is the result of performing the same test on my second account, where those champs are all unawakened but same rank:
The (actual) PI of the first team is 15863 (this includes sig level, masteries, etc). The actual PI of the second team is 13306. If the game was matching based on actual PI, I would expect the matches for the second account to be noticeably lower - by about 16% on average. Instead, they are very similar. If I overlap the two:
For all intents and purposes, that's overlapping, which is extremely strong evidence that the game is in fact using nakedPI for at least a significant component of match making.
Case closed, right? Well, maybe. I should point out there's another bit of strong supporting evidence for the nakedPI theory, and that is that if you go back and recalculate all of the actual fights in terms of nakedPI and compare the difficulty (as measured by the ratio of OpponentNakedPI / PlayerNakedPI) you get a much smoother monotonically increasing curve:
At least until the Kang matches show up, which in this case happens in round 8. The question is: if the game is using NakedPI to perform match calculations, why is this obfuscated in my own testing? Specifically, why did my testing, which concentrates on reverse calculating multiplier ratios from death match triggering thresholds, work at all? My best guess at the moment is that this is an emergent behavior of the arena. The arena is programmed to send matches at the player by an algorithm that in part observes what players actually use, and that implicitly also incorporates what players rank and how they sig up champions. It is possible that the way I've observed the arena is how it *happens* to work, at least for me and anyone like me, and the underlying math is different but drives the arena towards the behaviors I've observed so far.
The next step I think is to repeat my testing in two ways: first, take the same teams and rerun the arena streaks but always pulling the middle and the high matches, to see how they differ from the easy match. And second, to find a set of teams that, when measured by nakedPI, would allow me to redo all of my death match testing to compare to the original results in a way that would be distinguishing.
That may take some time, but I'll keep looking at it as time permits. This isn't the end of this particular discussion, because nakedPI brings up a very interesting question. If the game is using *only* nakedPI to choose opponents in the arena, does that mean by random chance some players have a significant advantage in the arena? Consider: suppose when I bring team X into the arena the game wants to find a nakedPI match of say 20000 PI. Suppose then that in my search space of matches the game finds a particular team that has a nakedPI of 20k, and in @CoatHang3r 's account the game also finds the exact same team composition, so it also has a nakedPI of 20k. But mine is max sig and in CoatHang3r's match they are all unawakened. The game considers both those situations to be the same situation, but since my match has higher *visible* PI, I will score more points beating that team than CoatHang3r will beating his team. That seems ... odd. Not impossible, but something to consider, if it turns out this is the whole answer. And remember that random chance can't solve this problem by evening things out, because the operating theory is that every player is pulling their matches from some subset of all players. If you get stuck with a bad subset of players that just don't like sigging up their champs, you could be stuck with that situation for a significant amount of time.
On BG’s stream when the new arenas first came out, one Kabam Dev did say that they can eliminate death matches if they wanted to. Not sure why they haven’t proceeded with that.
On BG’s stream when the new arenas first came out, one Kabam Dev did say that they can eliminate death matches if they wanted to. Not sure why they haven’t proceeded with that.
Because this is not a simple situation for two separate reasons. First, there's the question of what you define a death match to be, and second there's the question of what side effects happen when you attempt to eliminate them.
The maximum difficulty fight the game tries to send to you is about 4x you (whether that's actual PI or naked PI we'll set aside for now). What happens if we decide to define a death match to be any match higher than 2x, and just tell the game to lower the maximum from 4x to 2x, which basically eliminates such matches. Better, right?
Nope. The problem here is that players have gotten used to "infinite streak." But infinite streak is not the arena working correctly, it is the arena being broken. It can't find the matches it wants, so it throws easy matches at you. It can't find what it wants, because it is being asked to find very high matches - 3x, 4x, maybe even higher. So what happens if you lower that cap to 2x? Now, matches where currently the game can't find a match, it now will be able to find a match. Whereas now you're getting easy matches, you'll start getting 2x matches. The arena gets *harder* now. This is actually what happened the first time they tried to lower arena death match difficulty. The worst case death match got easier, but people started getting that all the time, which made the arena overall much harder, and much more stressful.
One way to look at it is that given how the arena works now, the harder the maximum difficulty is, the easier it is to break the arena and get infinite streak. So basically, the easier it is to get infinite streak, the worse the punishment is for losing it. Conversely, the lower the punishment is, the harder it is to get infinite streak in the first place. They are inversely connected. To change this you'd have to change the mechanics of the arena, which is more complicated.
Why not just lower the difficulty all the way down to 1x, then? Make it so that there's no difficulty curve at all, and therefore no way to make or break a streak? Well, because that's also very tricky. Originally I thought that was not possible because of random factors in the search algorithm. But CoatHang3r's NakedPI theory creates a separate issue: the PI you see is not the PI the game is using to generate matches. Either PI will still be potentially much higher than the player in many cases (especially for players that are not using masteries like suicides) or we have to factor the worst case scenario and lower the actual multiplier to even less than 1.0x, so the effects of masteries and signature levels do not inflate the champions beyond the threshold of 1.0 or so.
But that will now make the arena matches so easy that you could do them practically with your eyes closed most of the time. And that creates a new problem: it makes the arena even easier to bot. Bots are already a major problem in the arena; making the arena fights easier to automate could magnify the problem even further. It could even make it harder to identify bots, because bots do not fight like humans do, but in an arena where the fights are too easy it can become more difficult to tell the difference.
So they can eliminate death matches in a number of ways. But doing so has side effects that can be worse than the death matches themselves. So that's something that still lacks a perfect solution so far.
Comments
But the fact that your infinite streak system seems to be based around the notion that going below 15k is dangerous is consistent with my analysis that says going below 14550 is dangerous, within the fuzzy limits of precision I can currently claim.
*Actually I’m running suicides now but haven’t run them regularly for quite some time and even prior I never hit a death match running the bottom of my roster based on my naked pi. The bottom of my roster with suicides is 15-16.6k which drops significantly without suicides and I want to say well below the 14.6k.
For this to be valid, Gambit has to show up as a competitive max rank max sig champ, meaning 5/50 sig 99 in the 4* arena or R3 sig 200 in the 6* arenas. The former might be easier to catch in the wild than the latter. This is not something I specifically looked out for, but I will be specifically looking for it now.
If doom commonly makes it to teams with other high naked pi champs and gambit makes it to teams with lower naked pi it skews what will be returned for matches.
Another possibility: since different players do not tend to get the exact same matches, but one player grinding a lot can often notice seeing the same guys over and over again, this suggests that the matches I get and the matches you get aren't coming from some gigantic random pool. Instead we are each seeing some smaller, maybe hash table subset of all possibilities that could be different between us (that may also explain why in half a million fights of arena grinding I've never matched against myself). So it is possible that there is some small variations in what is possible for me and what is possible for you, and it is also possible this may change over time. So what's "safe" and "not safe" might be blurred by these additional dynamic factors.
This may be something that will need to be characterized over a lot of testing.
The difference between the two? The first was less than 25% of the max naked pi available in arena (gambit, doom, surfer) and the second was greater than 25% of the max naked pi.
When I do the math for the team that did, and did not trigger deathmatch I get 2966 (Falcon 4/40) + 2699 (Gamora 4/40) + 2862 (Ultron 4/40) = 8527 as the naked PI for the team that did not trigger death match and 2957 (DDHK 4/40) + 2915 (Sasq 4/40) + 2914 (Mr F 4/40) = 8786. Both teams appear to have higher naked PI than 25% of the maximum naked PI (which would be 8005).
Also, the naked PI of the death match team you did pull is (if I understand naked PI correctly) 30152. That's about 3.54x the naked PI of the team you used. That seems odd, if I understand your naked PI theory correctly this team should have been closer to 4x your own team's naked PI.
What am I doing wrong with regard to computing naked PI value?
Maybe the modifier floats as you suggested before. Or there is a slight rank/rarity modifier in addition to the 4x. However eliminating low naked PI from my r2 5s prevented me from ever encountering a death match while being able to mindlessly throw those champions into the arena. The 5* team of 13760 I showed has been routinely ran through arenas with zero death matches, even without suicide masteries which give it an even lower PI rating.
*the team in that picture isn’t the highest available team pi, it’s the lowest by about 11k, my options were NF or Elsa led teams (I have no idea who was paired with those teams, more on this in the spoiler) and doom was the easier match up so I choose him. The game obviously doesn’t solely choose only 4x teams but rather pulls from a variety of ~4x teams.
There are two things I’ve learned from arenas In relation to death matches. Minimum x PI doesn’t exclude you returning a death match, and ranking from the lowest naked PI will eventually eliminate the possibility of returning a death match when it comes to 5*.
Well maybe 3 things because the lower you go with naked pi the higher the chance of finding a death match, also the inverse, but that’s related to ranking the lowest naked pi. This was born of observing and analyzing countless death matches where the only common denominator was naked PI irregardless of displayed PI; pairing 2 low naked PI champs with the same rank/level could return a death match and only pairing 1 would not return a death match, additionally pairing 3 higher naked PI and low displayed PI wouldn’t return a death match. This to me eliminated displayed PI as a factor.
I choose the elsa team this time so show what was behind it. Naked pi was elsa-10002, domino-9853, doom-10434
I will run the same team when they come off cooldown and record the numbers and possibly sig level this time for true prestige.
I also fought 6 other matches testing the plain 4* naked PI and returned two death matches contrary to my straight naked PI theory. Being this works as 5* and not as 4* at the same challenger rating suggests there might be and additional value at play.
We can't sig up and down champions, but I can in theory perform an analogous test by comparing two different accounts with champions at max sig vs unawakened. I'm not sure if I will have the time to perform this test this weekend as I might be tied up, but I'll throw this out there if this is something you want to tackle yourself (or anyone else wants to do so).
And as it turns out, that test proved worth the units:
This is the results from my main account, using the team of 4* Gwenpool 5/50 sig 99, Morningstar 4/40 sig 99, and Immortal Ironfist 4/40 sig 99. The three curves are the PI I faced in each arena round. And this is the result of performing the same test on my second account, where those champs are all unawakened but same rank:
The (actual) PI of the first team is 15863 (this includes sig level, masteries, etc). The actual PI of the second team is 13306. If the game was matching based on actual PI, I would expect the matches for the second account to be noticeably lower - by about 16% on average. Instead, they are very similar. If I overlap the two:
For all intents and purposes, that's overlapping, which is extremely strong evidence that the game is in fact using nakedPI for at least a significant component of match making.
Case closed, right? Well, maybe. I should point out there's another bit of strong supporting evidence for the nakedPI theory, and that is that if you go back and recalculate all of the actual fights in terms of nakedPI and compare the difficulty (as measured by the ratio of OpponentNakedPI / PlayerNakedPI) you get a much smoother monotonically increasing curve:
At least until the Kang matches show up, which in this case happens in round 8. The question is: if the game is using NakedPI to perform match calculations, why is this obfuscated in my own testing? Specifically, why did my testing, which concentrates on reverse calculating multiplier ratios from death match triggering thresholds, work at all? My best guess at the moment is that this is an emergent behavior of the arena. The arena is programmed to send matches at the player by an algorithm that in part observes what players actually use, and that implicitly also incorporates what players rank and how they sig up champions. It is possible that the way I've observed the arena is how it *happens* to work, at least for me and anyone like me, and the underlying math is different but drives the arena towards the behaviors I've observed so far.
The next step I think is to repeat my testing in two ways: first, take the same teams and rerun the arena streaks but always pulling the middle and the high matches, to see how they differ from the easy match. And second, to find a set of teams that, when measured by nakedPI, would allow me to redo all of my death match testing to compare to the original results in a way that would be distinguishing.
That may take some time, but I'll keep looking at it as time permits. This isn't the end of this particular discussion, because nakedPI brings up a very interesting question. If the game is using *only* nakedPI to choose opponents in the arena, does that mean by random chance some players have a significant advantage in the arena? Consider: suppose when I bring team X into the arena the game wants to find a nakedPI match of say 20000 PI. Suppose then that in my search space of matches the game finds a particular team that has a nakedPI of 20k, and in @CoatHang3r 's account the game also finds the exact same team composition, so it also has a nakedPI of 20k. But mine is max sig and in CoatHang3r's match they are all unawakened. The game considers both those situations to be the same situation, but since my match has higher *visible* PI, I will score more points beating that team than CoatHang3r will beating his team. That seems ... odd. Not impossible, but something to consider, if it turns out this is the whole answer. And remember that random chance can't solve this problem by evening things out, because the operating theory is that every player is pulling their matches from some subset of all players. If you get stuck with a bad subset of players that just don't like sigging up their champs, you could be stuck with that situation for a significant amount of time.
So, worth continuing to investigate further.
The maximum difficulty fight the game tries to send to you is about 4x you (whether that's actual PI or naked PI we'll set aside for now). What happens if we decide to define a death match to be any match higher than 2x, and just tell the game to lower the maximum from 4x to 2x, which basically eliminates such matches. Better, right?
Nope. The problem here is that players have gotten used to "infinite streak." But infinite streak is not the arena working correctly, it is the arena being broken. It can't find the matches it wants, so it throws easy matches at you. It can't find what it wants, because it is being asked to find very high matches - 3x, 4x, maybe even higher. So what happens if you lower that cap to 2x? Now, matches where currently the game can't find a match, it now will be able to find a match. Whereas now you're getting easy matches, you'll start getting 2x matches. The arena gets *harder* now. This is actually what happened the first time they tried to lower arena death match difficulty. The worst case death match got easier, but people started getting that all the time, which made the arena overall much harder, and much more stressful.
One way to look at it is that given how the arena works now, the harder the maximum difficulty is, the easier it is to break the arena and get infinite streak. So basically, the easier it is to get infinite streak, the worse the punishment is for losing it. Conversely, the lower the punishment is, the harder it is to get infinite streak in the first place. They are inversely connected. To change this you'd have to change the mechanics of the arena, which is more complicated.
Why not just lower the difficulty all the way down to 1x, then? Make it so that there's no difficulty curve at all, and therefore no way to make or break a streak? Well, because that's also very tricky. Originally I thought that was not possible because of random factors in the search algorithm. But CoatHang3r's NakedPI theory creates a separate issue: the PI you see is not the PI the game is using to generate matches. Either PI will still be potentially much higher than the player in many cases (especially for players that are not using masteries like suicides) or we have to factor the worst case scenario and lower the actual multiplier to even less than 1.0x, so the effects of masteries and signature levels do not inflate the champions beyond the threshold of 1.0 or so.
But that will now make the arena matches so easy that you could do them practically with your eyes closed most of the time. And that creates a new problem: it makes the arena even easier to bot. Bots are already a major problem in the arena; making the arena fights easier to automate could magnify the problem even further. It could even make it harder to identify bots, because bots do not fight like humans do, but in an arena where the fights are too easy it can become more difficult to tell the difference.
So they can eliminate death matches in a number of ways. But doing so has side effects that can be worse than the death matches themselves. So that's something that still lacks a perfect solution so far.