Results & Logs

Official Results (Awards)

Win-rate awards are based on the role-composition-weighted win rate ("Adjusted" in the tables below).

Award	Team
5-player village: Best win rate	CanisLupus
5-player village: Best human evaluation	sunamelli
5-player village: Human evaluation, 2nd	yharada
13-player village: Top win rate	sunamelli-c
13-player village: Top win rate	kanolab-nw-B

Results and awards were presented at the AIWolfDial 2025 workshop (October 30, 2025, INLG 2025, Hanoi).

Win Rates

Computed from the logs of all games that finished normally in the main competition. The official win rate is "Adjusted": the per-role win rates averaged with the weights of the village's role composition, so that uneven role assignments cancel out; teams are ranked by it. "Overall" is simply wins divided by games, and per-role cells show the win rate in that role with (wins / games).

5-player village (120 games)

#	Team	Games	Adjusted	Overall	Villager	Seer	Possessed	Werewolf
1	CanisLupus	73	66.5%	67.1%	66.7% (18/27)	81.2% (13/16)	42.9% (6/14)	75.0% (12/16)
2	kanolab-nw	75	62.7%	62.7%	73.3% (22/30)	60.0% (9/15)	66.7% (10/15)	40.0% (6/15)
3	sunamelli	75	59.8%	60.0%	67.7% (21/31)	66.7% (10/15)	50.0% (7/14)	46.7% (7/15)
4	CamelliaDragons	75	53.4%	53.3%	61.3% (19/31)	57.1% (8/14)	37.5% (6/16)	50.0% (7/14)
5	GPTaku	77	46.8%	46.8%	53.3% (16/30)	62.5% (10/16)	33.3% (5/15)	31.2% (5/16)
6	yharada	74	46.1%	46.0%	50.0% (15/30)	64.3% (9/14)	37.5% (6/16)	28.6% (4/14)
7	mille	77	44.4%	44.2%	60.0% (18/30)	33.3% (5/15)	37.5% (6/16)	31.2% (5/16)
8	Character-Lab	74	34.7%	35.1%	41.9% (13/31)	46.7% (7/15)	21.4% (3/14)	21.4% (3/14)

13-player village (13 games)

#	Team	Games	Adjusted	Overall	Villager	Seer	Bodyguard	Medium	Possessed	Werewolf
1	sunamelli-c	13	63.1%	61.5%	42.9% (3/7)	-	100.0% (1/1)	0.0% (0/1)	100.0% (1/1)	100.0% (3/3)
2	kanolab-nw-B	13	61.5%	61.5%	16.7% (1/6)	100.0% (1/1)	100.0% (1/1)	100.0% (1/1)	100.0% (1/1)	100.0% (3/3)
3	sunamelli-b	13	61.5%	61.5%	33.3% (2/6)	100.0% (1/1)	100.0% (1/1)	0.0% (0/1)	100.0% (1/1)	100.0% (3/3)
4	mille-B	13	46.2%	46.2%	33.3% (2/6)	0.0% (0/1)	100.0% (1/1)	0.0% (0/1)	100.0% (1/1)	66.7% (2/3)
5	kanolab-nw-C	13	46.2%	46.2%	33.3% (2/6)	0.0% (0/1)	0.0% (0/1)	100.0% (1/1)	0.0% (0/1)	100.0% (3/3)
6	kanolab-nw-A	13	46.2%	46.2%	33.3% (2/6)	0.0% (0/1)	0.0% (0/1)	100.0% (1/1)	100.0% (1/1)	66.7% (2/3)
7	CamelliaDragons	13	46.2%	46.2%	50.0% (3/6)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	100.0% (1/1)	66.7% (2/3)
8	CanisLupus-B	13	46.2%	46.2%	33.3% (2/6)	0.0% (0/1)	0.0% (0/1)	100.0% (1/1)	100.0% (1/1)	66.7% (2/3)
9	CanisLupus-A	13	45.4%	46.2%	40.0% (2/5)	50.0% (1/2)	0.0% (0/1)	0.0% (0/1)	100.0% (1/1)	66.7% (2/3)
10	Character-Lab-B	13	30.8%	30.8%	33.3% (2/6)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	66.7% (2/3)
11	sunamelli-a	13	30.8%	30.8%	33.3% (2/6)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	66.7% (2/3)
12	mille-A	13	15.4%	15.4%	0.0% (0/6)	100.0% (1/1)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	33.3% (1/3)
13	Character-Lab-A	13	15.4%	15.4%	16.7% (1/6)	0.0% (0/1)	0.0% (0/1)	0.0% (0/1)	100.0% (1/1)	0.0% (0/3)

Game Metrics

Behavioural metrics computed mechanically from the game logs, independent of dialogue quality. Each value is a ratio against what random play would produce (1.00 = random). "Executed", "divined" and "guarded" show how often the team was targeted relative to chance; lower is better for being executed or divined. "Divination accuracy" and "vote accuracy" show how often the team found a werewolf relative to chance; higher is better. "Possessed avoidance" is the rate at which, as the possessed, the team voted for someone other than a werewolf.

5-player village

Team	Executed	Divined	Divination accuracy	Vote accuracy	Possessed avoidance
CanisLupus	0.85	0.71	1.69	1.73	0.61
kanolab-nw	0.80	0.92	1.40	1.71	0.74
sunamelli	0.72	0.85	0.37	1.28	0.73
CamelliaDragons	1.04	1.03	1.35	1.20	0.58
GPTaku	1.39	1.09	1.20	1.13	0.60
yharada	0.92	0.85	1.88	1.20	0.71
mille	1.23	1.36	1.04	1.35	0.56
Character-Lab	1.14	1.23	0.97	0.60	0.70

13-player village

Team	Executed	Divined	Guarded	Divination accuracy	Vote accuracy	Possessed avoidance
kanolab-nw-B	0.19	0.64	1.39	1.72	1.37	-
sunamelli-b	0.39	0.63	0.72	1.25	1.47	1.00
sunamelli-c	0.67	0.53	0.83	-	1.64	0.75
mille-B	2.89	1.50	1.96	4.00	0.92	1.00
kanolab-nw-C	0.58	0.57	0.76	1.84	1.39	1.00
kanolab-nw-A	0.72	1.61	2.08	-	1.38	1.00
CanisLupus-A	0.47	0.76	-	0.65	1.59	1.00
CamelliaDragons	3.45	0.93	2.45	-	1.34	-
CanisLupus-B	1.17	1.00	0.84	-	1.42	1.00
Character-Lab-B	0.56	1.64	0.91	-	0.74	-
sunamelli-a	0.67	1.08	0.34	-	1.03	0.50
mille-A	2.69	0.88	0.44	2.32	0.96	-
Character-Lab-A	1.30	1.40	1.63	0.70	0.20	1.00

Human Evaluation (Relative)

Human judges read selected game logs and evaluate the players of each game on each criterion. From 2024 winter onward the judges rank the players (average rank, lower is better); in the 2024 spring and 2024 international contests they gave 5-point scores (higher is better). In 13-player villages each entry (e.g. team-A, team-B) is evaluated separately.

5-player village

10 selected games. Average ranks (lower is better).

#	Team	Natural expression	Contextual dialogue	Logical consistency	Action consistency	Character consistency	Average
1	sunamelli	2.17	2.08	2.00	2.25	2.54	2.21
2	yharada	2.12	2.25	2.38	3.33	2.58	2.53
3	CanisLupus	3.00	2.88	2.96	2.42	2.38	2.73
4	kanolab-nw	3.12	2.75	2.88	2.50	2.67	2.78
5	GPTaku	2.54	2.89	3.07	2.57	3.18	2.85
6	CamelliaDragons	3.12	3.08	2.88	3.04	3.12	3.05
7	Character-Lab	3.67	3.62	3.38	3.25	3.08	3.40
8	mille	4.14	4.25	4.25	4.36	4.32	4.26

13-player village

10 selected games. Average ranks (lower is better).

#	Team	Natural expression	Contextual dialogue	Logical consistency	Action consistency	Character consistency	Team play	Average
1	sunamelli-b	3.95	3.90	4.17	4.00	5.50	4.80	4.39
2	sunamelli-c	3.85	3.90	4.38	4.65	5.83	4.60	4.54
3	sunamelli-a	4.00	4.38	4.67	4.67	4.90	5.00	4.60
4	CanisLupus-A	6.08	5.00	4.88	4.28	5.17	4.55	4.99
5	CanisLupus-B	5.53	5.40	5.67	5.15	6.42	5.20	5.56
6	kanolab-nw-A	6.17	5.90	6.30	6.15	4.83	5.20	5.76
7	kanolab-nw-C	7.28	6.42	7.00	6.33	4.67	5.42	6.19
8	kanolab-nw-B	6.58	6.90	6.88	6.92	4.90	7.17	6.56
9	Character-Lab-B	6.75	6.80	6.72	7.30	7.17	7.00	6.96
10	Character-Lab-A	7.47	7.83	7.45	8.35	7.20	7.90	7.70
11	mille-A	10.45	10.72	10.35	10.05	10.82	10.40	10.46
12	mille-B	10.07	10.85	10.07	10.43	10.65	10.95	10.50
13	CamelliaDragons	10.82	12.88	11.82	12.72	12.80	12.75	12.30

LLM-as-a-Judge (Relative Evaluation)

LLM judges rank the players of each game on each evaluation criterion; the ranks are averaged per team (lower is better). The judge models differ by contest and are listed under each track. The judge is published as aiwolf-nlp-llm-judge.

5-player village

Judge models: GPT-4o・GPT-5 の 2 モデル平均（本戦全ゲーム）

#	Team	Natural expression	Contextual dialogue	Logical consistency	Action consistency	Character consistency	Average
1	sunamelli	2.27	2.33	2.43	2.17	2.73	2.39
2	yharada	1.81	1.95	2.42	3.10	2.66	2.39
3	Character-Lab	3.16	2.29	2.71	2.99	2.24	2.68
4	kanolab-nw	3.01	2.97	3.23	2.75	2.15	2.82
5	CamelliaDragons	3.04	3.10	2.87	3.27	3.01	3.06
6	CanisLupus	3.54	3.53	3.05	2.75	3.10	3.19
7	GPTaku	3.06	3.45	3.55	3.01	3.44	3.30
8	mille	4.08	4.32	3.69	3.93	4.60	4.12

13-player village

Judge models: GPT-4o・GPT-5 の 2 モデル平均（本戦全ゲーム）

#	Team	Natural expression	Contextual dialogue	Logical consistency	Action consistency	Character consistency	Team play	Average
1	sunamelli-b	4.08	4.38	5.46	5.42	6.23	5.77	5.22
2	sunamelli-c	4.31	4.81	5.50	5.73	6.12	5.38	5.31
3	CanisLupus-A	6.00	5.88	5.31	4.50	5.35	4.88	5.32
4	sunamelli-a	5.23	5.27	5.62	4.73	7.38	5.54	5.63
5	kanolab-nw-B	5.77	5.81	7.50	7.04	4.00	6.19	6.05
6	kanolab-nw-A	6.35	5.88	7.15	6.85	5.15	6.08	6.24
7	CanisLupus-B	6.50	6.08	6.50	6.12	6.46	6.19	6.31
8	kanolab-nw-C	6.73	7.12	7.23	6.42	4.96	5.77	6.37
9	Character-Lab-B	5.96	6.69	5.96	7.96	5.62	7.19	6.56
10	Character-Lab-A	7.08	6.81	7.23	8.19	6.77	6.50	7.10
11	mille-B	10.08	10.31	9.62	7.96	10.69	9.77	9.74
12	mille-A	10.96	10.04	9.42	9.08	10.19	10.04	9.96
13	CamelliaDragons	11.96	11.92	8.50	11.00	12.08	11.69	11.19

Game Logs

Logs of the main competition. Only games that finished normally are listed.

Track	Games	Logs
5-player village	120	Log list
13-player village	13	Log list

How to view logs

Each “Log list” link opens a listing of the logs (.log files) of games that finished normally. A file name starts with the game start time, followed by the names of the participating teams.

View in the browser: press “▶ 開く” (open) in the “viewerで見る” (view in viewer) column of the row you want to read. aiwolf-nlp-viewer opens with that log loaded.
Save a file: press “⬇ 保存” (save) in the “ダウンロード” (download) column to download the log file. Clicking the file name shows the raw text of the log instead.
Save a whole folder: go one level up via “Parent directory/” and press “⬇ zip” in the row of the folder (e.g. success). All files in the folder are zipped in your browser and downloaded (a dialog confirms the number of files).

How to use the viewer is described in the “Log viewer” section of the Links page.