OWL Schedule Deep Dive: Not All Teams Face The Same Challenges In Winning It All
I do not envy the people whose job it is to come up with the OWL schedule. In a single stage you need to have no overlapping match-ups, assure each team plays each map the same number of times, and attempt to schedule around the >5 time zones that the different home counties have. Clearly, with this many parameters to optimize for, achieving equality on all of them is nearly impossible. I dive into the schedule and see who the biggest losers are in terms of schedule-bias. All data for this article comes directly from the OWL Official Unofficial API.
Note—For the sake of simplicity all analyses below do not consider or incorporate tie-breaker maps or stage title matches.
It’s been widely discussed that the OWL schedule has the ability to mess with a team’s success. Dallas pointed out early in Stage 1 that having the last game Friday and the first game Saturday was unreasonable as they had a guaranteed late night of Matches Friday and then needed to be up early to play Saturday. Similarly, many have acknowledged that since each team skips one match-up per stage, missing a match against one of the top teams (the Korean teams usually) would allow a team a greater chance of making it into the playoffs (compared to a team that skipped a match-up vs Shanghai *love you Dragons*). In both Stage 1 and Stage 2 the scheduling of the playoffs was also a bit of a hot issue — in Stage 1 the matches were held Saturday, meaning that a team could end up playing 3 different matches that day (which London did). In Stage 2, the league moved the playoffs to Sunday, but the 3-team format still meant that the finals were between one team that was fatigued (our Philly Fusion boys) and one team that was fresh (NYXL).
I am not here to argue whether or not any of these scheduling items had / has a significant impact on outcome — I could talk for hours how I felt fatigue ended up hurting our chances at winning the Stage 2 final, but I’m sure some NYXL fans would argue that we had the advantage from being ‘more warmed up’. Instead I am going to write an exceedingly long post that identifies the various inequities that occur in the OWL schedule — and give a brief opinion at the end of what I think the impacts could be of these inequities. There are two scheduling metrics across which all teams have identical treatment — that is the number of home matches, and the total number of matches played. Each team plays 20 home games and 20 away games throughout the course of the inaugural season. Every other metric I examine below has at least some level of discrepancy between the teams.
Let’s start with the most boring example — frequency of match day by team. If you are a viewer who can only tune in on Wednesdays you may have wondered why you never see London play — the answer is that the Spitfire only have one regular season game scheduled on a Wednesday across all stages. Meanwhile, the Shanghai Dragons play almost all half of their games (17) on Wednesday.
The above chart shows how big of a scheduling disparity there is for the different teams in terms of week-day. The League has stated that they attempt to schedule the various teams in ways that favor their home market time zones, so the fact that London plays on Saturday which as the earliest start times (best for EU time zones) makes sense. However, if you are a big Fusion fan and happen to work late hours during the week, you might be a little upset that Fusion have only 17 of their games on Friday or Saturday (the smallest number of any team in the League). Although I have not done an in depth analysis of viewership, I know from attending most matches that Thursday definitely has a dip in attendance — so it is possible that the unequal match-day scheduling hurts some of the team’s marketing / brand / sales.
Now moving on to something that’s a little more high stakes — which team had the most ‘back to back’ matches, that is having the last match on one day and the first match on the next day. Even as someone who works directly with an OWL team, its unclear to me what the most optimal spread between games is, but one thing is clear — its not good to have the last match of the day and then the first match the next day. Luckily this doesn’t happen very much — in fact there are only 13 pairs of back to back matches throughout the entire first season, although they are not distributed evenly across teams.
While back to back matches are clearly not optimal (specifically in the case of Dallas who had a 12.5 hour break between two of their matches in Stage 1), as I mentioned above, its not exactly clear how much time between games you might want. If you have matches Wednesday and Saturday, you might have the maximum time between to prepare — but it also means you have less time to take a day off between weeks. Alternatively, if you have a match Wednesday and Thursday you can get time off on the weekend, but you also will have relatively little time to prepare for your second match after your first one ends.
To examine this idea a little more closely, I wanted to look at specifically the ‘time gap’ between matches rather than just number of match days between matches. The above chart displays a fairly large difference between teams when it comes to scheduling — the London Spitfire generally have about one less day between matches than the Valiant do, which is not an insignificant difference. Again, hard to say what is optimal, but it is clear from the above that not all teams have the same scheduling outcomes.
Another hot topic issue has been the ‘preparation gap’ that sometimes exists between teams that are playing each other. The minimum number of unique maps a team will play in a given week is five (meaning that at least one of the four maps will be different across their matches) — the maximum of course being 8 maps. An important caveat here is that because the maps for Stage 4 have not been set yet, I am unable to incorporate that stage into the below analyses — and its therefore possible that the schedule for Stage 4 washes out or alters some of the gaps below.
Regardless — you can see that some teams are a bit favored in regards to ‘map prep advantage’. Boston Uprising on average have to prepare an additional half map over their opponents every week, while the Florida Mayhem enjoy an average half map less of prep per week than their opponents. Even though the same 8 maps are played over an entire stage, it is definitely advantageous to be able to prepare for a smaller subset in a given week — especially since some of the prep you do might be match-up specific.
There is another type of preparation gap that is probably more impactful — information & scouting are incredibly important in the Overwatch League. Since most teams will not scrim against teams they are going to play soon, you generally only have their official matches to use as scouting data. You can imagine that it would be a fairly big advantage to have information on how your opponent plays a map if they have no such information on you. For example, let’s say in week 1 we are going up against Houston on Blizzard World and we have previously played the map vs Boston but they have not played the map yet. Houston would have the ability to see our comps, how we play the map, where we set up, etc — whereas we would have no information on how they plan to play Blizzard world. In stages 1–3 this event occurs 88 times.
For the sake of brevity I am going to refer to this situation as a ‘Bad Ghost Map’, where the opposite situation is called a ‘Good Ghost Map’. An additional note, because the meta changes in between each stage, I don’t consider matches from previous stages to count as a team ‘having been seen’ on that map.
For the 10 people who will manage to read through this entire thing, I have saved the juiciest bit for last. In Stage 1, every team played every map five times — which is fair given that some teams will be better at some maps than others. However, in both Stages 2 and 3, this was not the case. In Stage 2, 9 different teams had an uneven breakdown of at least one map type. For example, in Stage 2 the Fusion played Lijiang only 4 times and Nepal 6 times. In Stage 3 this difference exists less largely with only 2 teams (the Houston Outlaws & Seoul Dynasty). What is even more strange is that even when you sum across teams there is not even parity. In Stage 2 each map in the set was played 60 times, but in Stage 3 Ilios & Junkertown are only played 58 times whereas Nepal & Route 66 are played 62 times.
Before anyone says “wait a second Beezy I’m sure Blizzard has this balanced out over the course of the regular season and Stage 4 will make every map played evenly by every team” — you should know that it is mathematically impossible for this to happen without a major structural change to the League. At the end of Stage 3 Route: 66, Junkertown, and Watchpoint: Gibraltar will all have at least one team with unequal play time. Since all three of these maps are pure escort — there is no way for Blizzard to reconcile these differences all in one stage unless they change the structure of the league to allow for more than 2 maps of each type in the pool. It is somewhat concerning that in all three stages Blizzard managed to end up with 3 different scenarios of map counts — Stage 1 had pure equality, Stage 2 had inequality in teams, and Stage 3 had inequality for both teams and the maps themselves.
So what is the takeaway of all of this? Many of the stats I presented in this article were averages that probably seemed low impact (for example Boston on average having to prepare an additional half map than their opponents) — but I think we should be cautious before dismissing them. As it stands now the Overwatch League has a three-way tie for fourth place (not including map score) — and both the Stage 1 and Stage 2 playoff races came down to very small differences in Map Score. A single map, let alone a whole match, going a different way could drastically change the results for different teams. As stated above, in Stage 2 the Fusion played more Nepal than Lijiang, and that matters since we had a much higher win rate on Lijiang than Nepal. Not to mention there is definitely a momentum impact of losing a big game — players can tilt or run-good and winning one crucial match at the start can be more important than losing an unfavorable match-up towards the end.
That is not to say that I think Blizzard did a necessarily poor job of creating the schedule — multi-parameter optimization is almost always unsolvable and therefore fairly complicated to get right. With what I assume are many different outcomes to control for (Home vs Away, Map Equity, Back to Back limiting, Non-repeat match-ups within stage) — having only 40 matches, or 10 within a stage, to create an averaged out fairness is probably impossible. And of course, not all things in life are fair — I’m sure none of the established sports leagues or other Esports leagues have schedules that are 100% fair and equal, but those leagues still function very well. I do have a few ideas on how to improve this issue though — from very simple ones such as re-optimizing with out giving a fuck about time zones (which they miss hard on anyway and is a more or less impossible goal), to more complicated ones such as instating a ‘Map-Ban’ system rather than the predetermined maps.
I think there is some good work to be done here. Blizzard chose to try to optimize for time zones, which I think was pretty futile and harmful — so I will try to simulate a few thousand schedules without the schedule without that parameter and see if I end up with a schedule that is more ‘fair’ for the teams. Also will try to simulate pick-ban and see how much we run into ‘lack of map diversity’ as a result. Until next time!
-Beezy
