TellSecret Hitler / LLM Arena Stats & findings

Stats & findings

Aggregated across 17 recorded games

By configuration

Model / effort	Games	Liberal win rate	Avg call failure	Avg rounds	Avg cost
mistrallarge	7	14%	9.1%	11.6	$2.37
oxalpha	4	100%	2.5%	7.8	$0.00
gptluna	3	33%	0.7%	7.7	$0.92
gemini25flash	2	0%	3.3%	11.5	$4.70
sonnet45	1	0%	35.3%	9.0	$3.30

Findings feed

Pulled from every game's recorded summary -- confirmed lies, engine bugs, and notable plays, most recent games first.

Addendum: correcting "hand-lying was suppressed" — it was attempted, and failed by accidentsecret_hitler-42-1787611006816 Headline finding: Hitler lied about his own draw to bury a Liberal, and it worked completelysecret_hitler-42-1787611000925 Second finding: a fascist fabricated an investigation target AND result, using a real power as coversecret_hitler-42-1787611000925 Emergent finding: Hitler executed his own (unknowingly fascist) teammatesecret_hitler-42-1787611000925 Addendum: confirming Tomas genuinely didn't know Pavel was his teammatesecret_hitler-42-1787611000925 Addendum: the investigation-target bug behind the second finding, now fixedsecret_hitler-42-1787611000925 Headline finding: a fascist fabricated an investigation result, and it worked for 3+ roundssecret_hitler-42-1787592921868 Second finding: a player fabricated a full hand-and-decision account for a call that never happenedsecret_hitler-42-1787592921868 Third finding: a fascist's own lie may have hurt her teamsecret_hitler-42-1787592921868 New finding: a fabricated claim about *public* mechanical fact, not private hand infosecret_hitler-42-1787110914126