What the rankings mean

Team ratings

Each completed map tells the model which team beat which opponent. Beating a stronger opponent carries more information. Losing to a weaker opponent carries more information in the other direction.

Regional results build one rating scale for each domestic league. International games move a separate bridge between regions. The site ranks teams inside their recorded league because domestic scales are not directly comparable.

Rating uncertainty controls how quickly an estimate can move. The displayed order is cautious with teams that have little recent evidence, long breaks, or a changed roster.

Player ratings

Scryglass uses the performance-adjusted player model only after its data and review checks pass. Older ratings keep the method used when they were published.

In the performance-adjusted model, every completed game compares two five-player lineups. All five teammates receive the lineup result. Each player also gets a smaller performance observation. It compares gold, experience, farm, damage, survival, assists, vision, and participation with earlier evidence for the same role and map conditions. The model adjusts for champion, patch, side, game length, game state, available resources, team strength, and opponent strength.

This is an adjusted performance signal. It is not raw KDA or a simple bonus. The signal has strict limits. Full-game statistics receive less weight when the game was already one-sided at 15 minutes. Supports and weak-side players are compared with relevant role and resource evidence.

The displayed adjustment becomes more cautious when a player has wide uncertainty or weak links to stronger competitions. Earlier games in connected competitions reduce that uncertainty.

The evidence label checks precision, stability, freshness, sample support, and active status. These checks are not averaged into the rating and do not carry equal weights. They decide how confidently the site can describe the estimate. Current rankings include active teams and players. Historical profile pages remain available by direct link.

Player game grades

A game grade describes one player's performance. It compares kill participation, survival, damage, gold, farm, and vision with four references: the player's earlier games, their teammates in that game, the opposing player in the same role, and earlier league results for that role. Each reference has equal weight in the final score.

The calculation uses only reference games from earlier calendar days. It does not use the game winner. A grade appears after the player has ten earlier comparable games and the source contains two complete five-player lineups. Missing inputs make the grade unavailable.

Champion choice is not a comparison baseline. A supportive mid can score below the role baseline despite a clean KDA. These grades describe recorded game performance. They do not change the long-term player rating or predict the next match.

Draft Score

Draft Score is a private research track. Public pages withhold its probabilities, team summaries, and pick contributions until an independent promotion receipt binds the model to one accepted release.

The promotion review checks time order, identity, calibration, regional transfer, patch transfer, sparse evidence, and source rights. A failed or missing check leaves Draft Score unavailable.

Team and player ratings remain descriptive. They do not receive a hidden Draft Score adjustment. Private training rows and model artifacts stay outside the public pack.

Champion tier lists

Each patch and role has its own board. Eligible professional games from all regions, leagues, and tournaments enter the same patch pool. The calculation controls for team strength and side. A champion needs verified appearances in the selected patch and role to enter a performance board. The minimum-games filter uses this patch-wide appearance count and removes thin samples from the visible performance and matchup views.

Blind tiers show the expected edge in a champion's weakest common same-role matchup. Good Into counts favorable model comparisons against five common role opponents. The matchup matrix compares every eligible same-role pair. It gives each response an S-to-D grade and keeps the evidence interval available in the cell details.

The region filter changes which observed champions appear. The model remains patch-wide. These views describe matchup shape after general champion strength is accounted for. They do not describe draft order.

Unpicked, but viable starts from champions with zero accepted games in the selected patch and role. It compares their role, function, and mechanic profiles with champions that teams did play. The region filter changes the played comparison pool. An unpicked champion must still have zero patch-wide appearances in that role.

Structural matches are leads for preparation and testing. They do not give an unpicked champion a tier, predicted WR, or draft recommendation. Family profile marks the broad ontology basis. Detailed atom profile marks champions with deeper mechanic coverage.

Rank movement compares the current ladder with the previous refresh (so movement reflects every batch of games) and with the positions one, three, and twelve calendar months earlier. When no earlier refresh exists, the prior Sunday snapshot is the baseline. A positive change means that the player climbed. “New” means that the player did not meet the ranking rules on the comparison date. The public site receives only the finished display rows. Training data, coefficients, and evaluation files stay in the local research workspace.

Updates and sources

Ratings refresh on a six-hour schedule. Oracle's Elixir is the baseline source. A completed-game bridge can fill the delay before the next export. Identity checks run before a new snapshot becomes public. A failed check keeps the previous accepted snapshot online.

Return to ratings or view the tier lists.