About DubsRank

What this is

DubsRank ranks every Super Smash Bros. Melee doubles player with a rating built only from doubles results. Singles rankings are everywhere; doubles is its own game, with its own specialists, partnerships, and upsets, and it deserves its own leaderboard. The goal is a fair, data-driven picture of who actually wins doubles sets: individually, as teams, and over time.

What you can do here

DubsRank is more than a leaderboard. On the Events page you can find upcoming Melee doubles tournaments, filtered by online or offline, country and state, so you can see what's coming up near you. The same page lets you search and open every doubles tournament from the past decade, with full results, placings and every set that was played.

You can also look up any player or team to see their rating history, best wins and teammates, compare players or teams on the Head-to-Heads page, and explore character and pairing trends on the Stats page.

Running a tournament? The Seeding & Bracket Projections tool on the Events page takes a start.gg link and gives you projected seeds for every team, so you can seed a doubles bracket from real results. Commentating? Player and team pages, head-to-head records, best wins, character stats and tournament history are all here to pull facts and storylines from during a set.

How the ratings work

Every doubles set from start.gg and parry.gg is replayed in the order it happened. For now, the data covers results from 2015 to the present, so older history isn't counted yet. Each player carries their own rating, and a team's chance of winning a set comes from both partners' ratings. Wins against stronger teams count for more, and the model also accounts for things like tournament size and strength, and how new a player is. The same engine powers the player and team rankings, win-rate stats, achievements, and teammate recommendations on this site.

How accurate is it?

We measure accuracy with the Brier score, a standard way to grade win predictions: 0 would be perfect, and always guessing 50/50 scores 0.25. DubsRank's predictions score about 0.14. That's good (the model is well-calibrated and precise), but it isn't perfect, and it's not as sharp as the best singles models. In its current state, it's around 85% accurate at predicting who wins a set.

Doubles is harder to predict: every set involves four players and partnerships change constantly. The biggest hurdles for the algorithm are players and tournaments with little data. A player with only a handful of recorded sets, or an event full of unknown players, gives the model much less to go on.

Project+ (P+) and Project M (PM) singles rankings

Click to expand

DubsRank also ranks every Project+ singles player (and Project M, the mod P+ grew out of), at Project+ All-Time Singles Rankings. It uses the same rating engine as the Melee doubles rankings, built for one-on-one sets: every P+ and PM singles set on start.gg, Challonge and parry.gg is replayed in the order it happened, wins against stronger players count for more, and bigger and stronger tournaments matter more.

Results reach back to the early Project M years. Old brackets run on Challonge are included alongside start.gg, and a handful of early majors whose brackets only survive as pictures (like Genesis 2, Zenith and The Big House 2) were typed in by hand. Where only the top of a major's bracket survives (a Top 8 to Top 64), those sets count a little extra, since everyone in them got there by beating a bigger field.

The multi-pass rebuild

A normal Elo replay goes through every set once, in order. That has a blind spot: when someone beats a player early in that player's career, the win counts as a win over a newcomer, even if that newcomer later turned out to be one of the best in the game. The P+ rankings fix this by replaying every set four times. Each pass starts every player at the rating they finished the previous pass with, then replays all the results again, so every win and loss is re-weighed by how strong each opponent really turned out to be. Players with only a few sets are kept closer to the normal starting rating, so one good event doesn't count for too much.

This makes the ratings more accurate: early results are judged fairly, and a strong player who is new to the game doesn't have to spend dozens of sets climbing up from the bottom before their rating means anything. Tested honestly (seeding every 2024–2026 bracket using only results from before it), the multi-pass ratings seed P+ brackets more accurately than a single pass did. Each tournament on a profile still shows the rating that player had at the time; the current rating and the ranks use every pass.

Many P+ brackets are listed under Super Smash Bros. Brawl, so Brawl events titled the P+ way ("Brawl" in quotes, or naming P+ or Project M) are included, as are P+ and Project M brackets listed under Melee. Brackets hidden under other games are reviewed by hand before they count. The skill classes work the same as the Melee ones, with GrandMaster for the top 100 players of all time and Master for the rest of the top 1%. Every P+ player has a profile with their Elo history, best wins, achievements and head-to-heads, and the P+ versions of the Events, Head-to-Heads, Stats, History and Character pages cover P+ only.

Rivals 2 singles rankings

Click to expand

DubsRank also ranks every Rivals of Aether II singles player at Rivals 2 All-Time Singles Rankings. It works exactly like the Project+ rankings: every Rivals 2 singles set on start.gg and parry.gg is replayed in the order it happened, wins against stronger players count for more, and bigger and stronger tournaments matter more.

The multi-pass rebuild

A normal Elo replay goes through every set once, in order. That has a blind spot: when someone beats a player early in that player's career, the win counts as a win over a newcomer, even if that newcomer later turned out to be one of the best in the game. The Rivals 2 rankings fix this by replaying every set four times. Each pass starts every player at the rating they finished the previous pass with, then replays all the results again, so every win and loss is re-weighed by how strong each opponent really turned out to be. Players with only a few sets are kept closer to the normal starting rating, so one good event doesn't count for too much.

This makes the ratings more accurate: early results are judged fairly, and a strong player who is new to the game doesn't have to spend dozens of sets climbing up from the bottom before their rating means anything. Tested honestly (predicting each set using only results from before it), it predicts who wins a Rivals 2 set about 4% better than a single pass (by Brier score), and it seeds brackets more accurately too. Each tournament on a profile still shows the rating that player had at the time; the current rating and the ranks use every pass.

The skill classes work the same way, with GrandMaster for the top 100 players of all time and Master for the rest of the top 300. Every Rivals 2 player has a profile with their Elo history, best wins, achievements and head-to-heads, and the Rivals 2 versions of the Events, Head-to-Heads, Stats, History and Character pages cover Rivals 2 only.

Smash 64 singles rankings

Click to expand

DubsRank also ranks every Super Smash Bros. 64 singles player at Smash 64 All-Time Singles Rankings. It works like the Project+ rankings: every Smash 64 singles set on start.gg, Challonge and parry.gg is replayed in the order it happened, wins against stronger players count for more, and bigger and stronger tournaments matter more.

On Challonge, “Super Smash Bros.” is a catch-all label that many Ultimate, school and casual brackets are filed under too. So a Challonge bracket only counts when there are signs it’s really Smash 64: its title or organiser page says 64, it was played before Ultimate came out, an SSB64 Rank player entered it, or many of its players have start.gg Smash 64 results.

The multi-pass rebuild

A normal Elo replay goes through every set once, in order. That has a blind spot: when someone beats a player early in that player's career, the win counts as a win over a newcomer, even if that newcomer later turned out to be one of the best in the game. The Smash 64 rankings fix this by replaying every set four times. Each pass starts every player at the rating they finished the previous pass with, then replays all the results again, so every win and loss is re-weighed by how strong each opponent really turned out to be. Players with only a few sets are kept closer to the normal starting rating, so one good event doesn't count for too much.

This makes the ratings more accurate: early results are judged fairly, and a strong player who is new to the game doesn't have to spend dozens of sets climbing up from the bottom before their rating means anything. Together with the clean-up above, it predicts who wins a Smash 64 set about 3% better (by Brier score) in an honest test of 2024–2026 sets. Each tournament on a profile still shows the rating that player had at the time; the current rating and the ranks use every pass.

Support the project

DubsRank is a passion project. If you enjoy it and want to help keep it running and improving, you can support me here: paypal.me/meticu

Support on PayPal

Find me