So to be clear, your PRESUMPTION (and yes I’m calling it a presumption, because no, Tale, I’m not going to unquestioningly accept that every claim you make is de facto true or applicable to OW, no matter how much experience you have with other sorting models in other contexts. Indeed, you could be a Nobel prize laureate in this field and that still would have no bearing on your claims IN THIS CONTEXT on whether your conclusions are conjectural) is that it is NOT POSSIBLE that grind or engagement based mechanics exist in Overwatch?
To answer your question, “well designed” systems are predictive to reduce variability and accelerate the sorting process.
Not sure I agree here. The ladder ranking seems more like a point-in-time snapshot reflecting prior player interactions. Your statement here seems to apply more to MMR?
You cannot make statements like “it says such and such.” I’ve already stated that my conclusions about the matchmaker are inferences. Your statements here, however, seem to assert facts, claiming to know what it’s doing and trying to do, and so you take on a higher burden of proof. What you describe here is the working of a SBMM and not an EBMM. EBMM would take other values and variables into account, and would bring matches together based on factors beyond the assessment of raw skill.
This seems to defy logic. If the predictions are wrong, and those predictions directly affect matchmaking actions, then the predictions could and certainly will be determinative in some cases: file the result under “unintended consequences.” If I incorrectly believe to be Joe a great baseball player, and I put him on team A under that presumption, and he’s actually a bad player, that prediction/conclusion about Joe absolutely has consequences for both teams. The expression of skill of every player on each team IS affected by Joe’s performance while playing, and his placement on whichever team was a result of my prediction. My prediction for better or worse, certainly has consequences which could affect the outcome.
To agree with this conclusion, I’d have to agree with your premises, which as you can see above, I do not.
Your explanation makes a lot of assumptions. It also doesn’t address a phenomenon I see quite often in my games, which is loss streaks following win streaks. Considering the reliability of this phenomenon, I no longer consider it possible to be a statistical anomaly.
Also, unlike the GMAT where a more knowledgeable, higher percentile test taker still “wins” even though their score is considerably lower than it would’ve been on a non-adaptive test, a 90th percentile OW player suddenly has the same odds of winning their match as a 10th percentile player. We know in a random match, the 90th percentile player only has about a 10-15% chance of seeing a player on the other team who’s about as good or better than they are. Under the current model, that number is conservatively, what, 70-90%? That player should also expect to see a MUCH higher than normal chance of being grouped with a bad player – the BETTER his results – whereas this is not true for a 10th percentile player.
In scenario A) we take two top high school varsity basketball players, Steve and Alan, and these two players are playing on teams otherwise composed of 5th graders. By random drawing, Steve and Alan can either be on the same team or on opposing teams. Steve advances a grade every time he wins 5 matches (6th, 7th, and so on).
In Scenario B) we ensure that Alan, again, a player equal in skill to Steve, is always on the opposing team, playing against Steve each and every match until the experiment is concluded
In scenario B) we should expect Steve’s winrate to drop considerably, as he will lose many more matches.
In scenario A) we would expect Steve to have a relatively high win rate until he reached players of his approximate age and skill.
What am I missing here?
Yes, and GMAC’s goals are almost certainly very different than Blizzard’s goals, which would logically lead to the two organizations designing/wanting different systems. GMAC wants to sort the best students from the worst and to place them along a spectrum. Blizzard wants to make as much profit as possible which – does not require – cleanly and accurately or efficiently sorting players by skill. It only requires that they keep players playing for as long as possible. Extensive research has shown that players exposed to engagement based matchmaking play longer, are less susceptible to churn, and therefore show higher engagements metrics in every category that matters. So I’m not sure why you believe that GMAT algorithms are relevant to another industry with different goals and different business models. It seems logical that there might be some overlap, but it also seems logical that there would be considerable differences.
See Steve and Alan scenario above about win rate differential between random groupings and 50% groupings. You also cannot say that a random matchmaker “serves no competitive ranking purpose” since the process of winning and losing is obviously a stratification mechanism. Your presumption is also that there is no intentional grind for the purpose of engagement, even though it’s a known fact that the gaming industry commonly institutes grind mechanics to boost profits and engagement.
Yes or no question: is having a 1000 SR spread in matchmaking likely to increase grind?
Yes or no question: is having no SR restriction per role, such that one team can get plat tanks and silver healers, and vice versa, likely to increase grind?
Yes or no question: is there any correlation between EBMM and profit/engagement?
What is your evidence or what leads you to be believe that Blizzard employs no EBMM in its matchmaking algorithms?
The planet would’ve burned to a crisp by now if we operated under the assumption that a group of people (all with serious biases/predispositions towards one outcome over another) performing in a highly unscientific, unsupervised environment had discovered objective truths. There are so many ways that this “test” could’ve gone wrong or been fudged or proved inconclusive that it’s really not worth mentioning.