Glickman 1999 · 2001, in your browser

Who is really the strongest in your club — and how sure is that?

Paste the results of your club's games. Every player gets a Glicko-2 rating and the range the rating could honestly be in, next to plain Elo. Two players whose ranges overlap cannot be told apart yet, whatever the ladder says. Nothing is uploaded.

Results

Open a CSV

One game per line: date, player, player, result. The date may be left out or be a round number. Results: 1-0, 0-1, ½-½, a score such as 11-7, or the winner's name. Commas, semicolons or tabs.

Settings

Rating period τ (how fast volatility may change) Elo K-factor

Games inside one rating period count as played at the same time. Everyone starts at 1500 with a deviation of 350 and a volatility of 0.06, as in Glickman's note.

Ratings

95% range (rating ± 2 × deviation)rating

All the numbers

Player	Glicko-2	RD	Volatility	Games	W–D–L	Elo

What this is

Elo gives every player one number and treats it as known. Mark Glickman's system adds what Elo leaves out: a rating deviation (how uncertain the rating still is — large for a newcomer or someone who has not played for months, small for a regular) and, in Glicko-2, a volatility (how erratic the player's results are). A rating of 1700 ± 60 and a rating of 1700 ± 220 are different pieces of knowledge, and a club ladder that hides the second number ranks people it cannot yet rank.

Checked against the source. The calculation follows Glickman's own step-by-step note and reproduces its worked example — a 1500 player (deviation 200) who beats a 1400 and loses to a 1550 and a 1700 ends at 1464.05, deviation 151.52, volatility 0.05999. The note prints 1464.06 because it multiplies rounded figures; its table of g and E values and each step of its volatility iteration match to the last printed digit.
Rating periods. Glicko-2 rates a batch of games at once and works best with a fair number of games per player in each batch (Glickman suggests 10–15). The page picks the shortest period — day, week or month — that gives each player at least five, and otherwise rates the whole file as one period. You can override it under Settings.
Sitting out. A player who misses a period keeps the rating, and the range widens; it never grows past the newcomer's 350.
The 95% range is rating ± 2 × deviation, as in the note. It describes this method's uncertainty about the player's strength, given only the games you pasted.
Head-to-head uses the expected-score formula of the 1999 paper, which takes both players' uncertainty into account. It is an expected score — a win counts 1, a draw ½ — not a win probability.
Elo is computed game by game in file order with the K-factor you set, everyone starting at 1500.
Dates like 05/03/2026 are read day first (5 March).

Sources: Glickman, M. E. (1999). Parameter estimation in large dynamic paired comparison experiments. Applied Statistics 48(3), 377–394. · Glickman, M. E. (2001). Dynamic paired comparison models with stochastic variances. Journal of Applied Statistics 28(6), 673–689. · Example of the Glicko-2 system (revision of 22 March 2022).