What tournaments were taken into account for #PGRSF?
The #PGRSF was our largest data collection effort yet. It contains every offline Capcom Pro Tour tournament from 2016 to 2019 (235 in total), plus notable non-CPT SFV events Evo Japan, Red Bull Kumite, Frosty Faustings, Northeast Championships, and Kumite in Tennessee from 2016 to early 2020. In total, a massive 254 brackets were tracked, more than 2.5 times the number tracked for #PGRZ.
Taking in feedback from #PGRZ, events were not weighted based on entrant count. Instead, tournament tiers were assigned based on Capcom Pro Tour status, with the non-CPT events assigned relative to those:
How does the PGstats algorithm work?
Broadly, it runs an iterative calculation which determines the strength of each player based on their wins, losses and outplacings within the season. The data points it takes in are each player’s matches and final standings at each PGR tournament. The actual calculation is not available to the public (see below).
Placements: Players are not rewarded directly on the actual numerical value of their placing at a given tournament; rather, their placement score is derived based on the other players in attendance and whether they outplaced or were outplaced by them. The person who finishes 1st receives head-to-head points against everyone else who attended; the person who finishes 2nd receives points against everyone else except the first-place finisher, and so on. This means that a tournament’s tier is simply one factor in the number of points a player receives for winning an event, and not the entire story. A lower-tiered event with significantly harder competition will reward more points than a higher-tiered event with few strong opponents.
Each player is given a head-to-head score against all other players who attended at least one tournament in common. This score is based on the given players’ set count, with outplacements treated as partial wins (a win/outplacement at a larger tournament is worth more than one at a smaller event). From these, we generate an objective list of each player’s score or “strength", the value of a win against that player. Thus, if two players have identical overall records against the top 50 but one player’s wins have come against stronger opponents, that player will have a higher score and will be higher on the list.
The algorithm’s confidence factor was not used for #PGRSF; every eligible player competed at so many events across 4+ years that the algorithm did not need to attempt to correct for low attendance.
Why is the algorithm private?
There are two main reasons why the algorithm isn’t shared with the public:
What does the algorithm take into account?
The algorithm’s only inputs are the wins, losses, and placements at PGR tournaments of the players who attended.
Why doesn't a panel change the rankings at the end?
If you have a panel change the results of the algorithm at the end, then there’s no real reason to have an algorithm at all. The point of the algorithm is to be objective; introducing a subjective panel into the mix will leave you with the worst of both worlds.
Does the algorithm have any form of recency bias?
It does not. We have tested including some form of recency bias in the past, but determined that this resulted in the algorithm favoring whoever peaked at the last big event. Each PGR is meant to be a complete history of the era it reports on, not a Power Ranking representing the current state of the scene. This means it will contain strong players who have stopped playing and consistent players who have performed at a high level for a long time, regardless of those players’ more recent results.
I want to know more about #PGRSF methodology and its details.
PGRSF Administrative Team
Alex "Vayseth" Varga, Video Director
Andrew "PracticalTAS" Nestico, Statistician
Jack "Jackie Peanuts" Moore, Editor
Max "CurlyW" Wasserman, Fact-Checker and Editor
Keith "keiththehuman" Jackson, Graphic Design
