You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+58-48Lines changed: 58 additions & 48 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -134,53 +134,9 @@ To find the optimal _p_, we ran many simulations, updating ELOs daily and compar
134
134
135
135
The sweet spot is _p_ = **1.11**, giving reasonably good results in line with expectations.
136
136
137
-
## FAQ
138
-
139
-
_This section is still being created._
140
-
141
-
### What is the deviation?
142
-
143
-
The **deviation** indicates how accurate or uncertain your rating is. You’ll appear on the list of **active players** if your deviation drops below **60–70** (depending on your rating), and you'll be removed from the list if it rises above around **100**.
144
-
145
-
To lower your deviation, you need to provide **new information** about your skill level. Continuously beating much weaker opponents (which might boost your rating in other systems) doesn't help here — it's not new information. For example, you’re expected to win more than **99%** of the time against someone **1000 ELO** below you. So winning 20 games in a row against such an opponent won’t improve your rating — it might even **increase** your deviation due to the lack of meaningful data.
146
-
147
-
That’s one reason why Glicko-2 is hard to exploit. The best way to reduce your deviation is to play a variety of opponents **close to your skill level**.
148
-
149
-
---
150
-
151
-
### How many points do I gain or lose per game?
152
-
153
-
**Short answer:** It depends.
154
-
155
-
Games are submitted to the system in **daily batches**, not individually. Your rating is updated based on your **overall performance** for the day, rather than on a per-game basis. This method provides a more accurate and stable rating.
156
-
157
-
---
158
-
159
-
### I didn’t care about my rating before, but now I want to take it seriously. Can you reset it?
160
-
161
-
**No.** Your rating reflects your performance over roughly the **last 200–300 games**. Older games gradually lose their impact.
162
-
163
-
If you play **50 games a day for a week**, your rating will effectively reset itself through recent performance. The result will be the same as starting from scratch - just without creating a new account.
164
-
165
-
---
166
-
167
-
### How bad is a loss against a much weaker opponent?
168
-
169
-
A single loss isn’t a big deal — even if your opponent is far below you in rating. The expected win rate is never 100%, no matter how large the ELO gap is. So if you lose one game against someone **1000 ELO** below you, it's likely just the one game out of 100 that you were expected to lose anyway.
170
-
171
-
But the **second** loss is much more significant. That result suggests that your strength may have been overestimated, and the system will start adjusting your rating accordingly.
172
-
173
-
---
174
-
175
-
### How often is the list updated?
176
-
177
-
The list is updated once per day, with the daily cutoff at UTC+5.
178
-
Your rating — as shown in the all-time leaderboard — reflects your status at the end of the previous day based on that time.
179
-
180
137
## Usage on CnCNet
181
138
182
-
Since ELO computation is processed in batches - to ensure higher accuracy - rather than after each game, this application needs to run once per day.
183
-
The "ELO day" ends at UTC+5 for all players, so the best time to run it is between UTC+5 and UTC+6.
139
+
Since ELO computation is processed in batches - to ensure higher accuracy - rather than after each game, this application needs to run once per day. The "ELO day" ends at UTC+5 for all players, so the best time to run it is between UTC+5 and UTC+6.
184
140
185
141
During the build of the Docker container, this repository is cloned and the app is compiled.
186
142
@@ -190,6 +146,16 @@ During the build of the Docker container, this repository is cloned and the app
This will generate ELO for blitz, using environment variables only. It also includes the specified tournament game file from the repository.
199
+
This will generate ELO for blitz, using environment variables only. It also includes the specified tournament game file. Since a tournament file is only used for blitz and not updated anymore, the layout of a tournament won't be explained here. Just throw `gameoverlay.cpp` at an AI and it will provide you an example JSON.
200
+
201
+
202
+
## FAQ
203
+
204
+
_This section is still being created._
205
+
206
+
### What is the deviation?
207
+
208
+
The **deviation** indicates how accurate or uncertain your rating is. You’ll appear on the list of **active players** if your deviation drops below **60–70** (depending on your rating), and you'll be removed from the list if it rises above around **100**.
209
+
210
+
To lower your deviation, you need to provide **new information** about your skill level. Continuously beating much weaker opponents (which might boost your rating in other systems) doesn't help here — it's not new information. For example, you’re expected to win more than **99%** of the time against someone **1000 ELO** below you. So winning 20 games in a row against such an opponent won’t improve your rating — it might even **increase** your deviation due to the lack of meaningful data.
211
+
212
+
That’s one reason why Glicko-2 is hard to exploit. The best way to reduce your deviation is to play a variety of opponents **close to your skill level**.
213
+
214
+
---
215
+
216
+
### How many points do I gain or lose per game?
217
+
218
+
**Short answer:** It depends.
219
+
220
+
Games are submitted to the system in **daily batches**, not individually. Your rating is updated based on your **overall performance** for the day, rather than on a per-game basis. This method provides a more accurate and stable rating.
221
+
222
+
---
223
+
224
+
### I didn’t care about my rating before, but now I want to take it seriously. Can you reset it?
225
+
226
+
**No.** Your rating reflects your performance over roughly the **last 200–300 games**. Older games gradually lose their impact.
227
+
228
+
If you play **50 games a day for a week**, your rating will effectively reset itself through recent performance. The result will be the same as starting from scratch - just without creating a new account.
229
+
230
+
---
231
+
232
+
### How bad is a loss against a much weaker opponent?
233
+
234
+
A single loss isn’t a big deal — even if your opponent is far below you in rating. The expected win rate is never 100%, no matter how large the ELO gap is. So if you lose one game against someone **1000 ELO** below you, it's likely just the one game out of 100 that you were expected to lose anyway.
235
+
236
+
But the **second** loss is much more significant. That result suggests that your strength may have been overestimated, and the system will start adjusting your rating accordingly.
237
+
238
+
---
239
+
240
+
### How often is the list updated?
241
+
242
+
The list is updated once per day, with the daily cutoff at UTC+5.
243
+
Your rating — as shown in the all-time leaderboard — reflects your status at the end of the previous day based on that time.
0 commit comments