You open the Renju ratings and notice someone near the top. Then you see that this player has few recent games and a conspicuous ± beside the score. Copying down the rank would miss half the table. To read a WHR rating, first check the update date and filters. Then follow the row through rating, game counts and uncertainty before returning to the actual matches.
First, Check When the Table Was Updated
A rating table is a dated snapshot. The Renju WHR homepage lists September 13, 2026, as its latest update and reports 5,563 players and 133,209 games. That is the date shown on the page; it does not prove that every match played by then has been included. The WHR method itself was described in a paper published in 2008.
The 5,563 players and 133,209 games describe the size of the dataset used by the page. They do not mean that the default table has 5,563 rows, or that every rated player has a recent record. Filters reduce the number of players you can see.
Next, inspect the controls: historical date, country or region, and Active, Gy1≥10 or All. The default view is Active. When taking a screenshot or comparing two players, record both the update date and the filter. Otherwise, the same rank may be impossible to reproduce.
Read the Rating Before the Rank
Start each row by checking the name and country or region, which helps prevent confusion over similar names or secondhand references. Then read Rating and ±, followed by Gy1, Gy2 and Gy5. Consider Rank last: it is simply the order produced when the current page sorts the selected group by rating.
A Rating is the model’s estimate of that player’s strength at the relevant point in time. It is useful for comparisons made within the same dataset, version and filtering context. There is no universal conversion to conventional ranks, and the number should not be matched digit for digit with a rating from another website.
Rank can change because ratings move, records are added or the filters change. If someone is tenth on one page and twelfth on another, first check whether the view changed from Active to All or whether a different historical year was selected. Only then should you consider a change in playing strength.
Gy1, Gy2 and Gy5 Count Recent Games
Gy1, Gy2 and Gy5 count a player’s games within the past one, two and five years, respectively. They are game totals, not wins, points or winning streaks. A high Gy1 means only that more recent games entered the dataset.
The published code that generates the table shows that Active requires at least 10 total games and a Gy5 greater than zero. Gy1≥10 further requires at least 10 games in the past year. All removes the activity filter, so the default page does not include every rated player.
Consider a hypothetical example: Player A and Player B have similar ratings. A has a Gy1 of 30, while B has a Gy1 of zero. You can say that A has a fuller recent record. You cannot conclude that A has performed better lately, because these columns do not show the balance of wins and losses.
Read the filter first, the rating second and the game record last.
What the ± Beside a Rating Can Tell You
The ± beside a rating is the table’s uncertainty field. A larger value can be read cautiously as a looser current estimate; a smaller value suggests that the model’s estimate is more concentrated. Its purpose is to slow down a quick conclusion.
The page does not assign a stated confidence level to this ±, so do not describe “2800 ± 80” as a 95 percent confidence interval. Whether two apparent ranges overlap is also insufficient to determine the precise probability of either player winning a single game.
It is safer to read ± alongside the Gy columns. When ratings are close, allow more room for doubt if one row has fewer recent games and a larger ±. An Active player can still have a Gy1 of zero because that filter asks whether the player has a record within five years, not whether the player competed in the past year.
A Historical Curve Is a Retrospective Estimate
The original WHR paper page explains that the method uses a dynamic Bradley–Terry model and the full history of the included data to estimate how a player’s strength changes over time. Technically, it finds the posterior mode across that history rather than simply averaging annual win rates.
A historical-year page is therefore closer to “an estimate of that moment using the dataset as it stands now.” When games are added, corrected or newly recorded, the model reuses the full history, and estimates for earlier dates may change. When citing an old rating, include the date on which you consulted the table.
This distinction is especially useful after a player has been inactive for a long time. Current, a selected historical year and the player’s last competition answer different questions: how the model places the player now, what the estimated trajectory looked like then, and what actually happened in that event. Do not compress all three into one judgment.
Ratings Add Context; Game Records Decide Results
WHR can provide context for a pregame comparison, but it cannot rewrite a match result. If the higher-rated player loses, the official event record still determines the outcome. Nor does one upset automatically invalidate an entire long-term rating table.
When examining a particular pairing, first record both players’ Rating, ± and Gy1 under the same update date. Then check the event date, rules, move order and game record. The rating gap describes the model’s background estimate; the moves reveal how the result developed, including where an open three, four-in-a-row or forbidden-move ruling appeared.
Make the note specific: “I am citing the Active ranking on the page updated September 13, 2026; recent activity appears in the Gy1 column.” If you selected a historical year, add that year. This qualification gives the quoted rank a source that another reader can check.