← mafiabench

$ replay · you see everything

s2-game-012

[ raw json ]view

Ready

0 / 227

// Transcript

click a seat to follow one model
  1. Events appear here as the replay advances.
// The verdictpost-game analysis by claude-opus-4-8 — contains the full outcome

Fabricated Deaths and Identity Slips Sink the Mafia

Sheriff Grok 4.20R opened a claim war by naming Fable 5 as Mafia, and Fable's reckless counterclaim collapsed under two mechanical lies — a 'refused twice' fabrication and offering a behavioral off-ramp to his own 'confirmed guilty' — letting town execute him Day 1. Day 2, Sonnet 5 torched himself by hallucinating a night kill that never happened, which G3.1 Pro and others instantly flagged, cementing the Sheriff-Angel core once Opus confirmed. G3 Flash's persistent 'coordinated bloc' narrative fooled Angel Opus into hunting townsfolk, and town even mislynched townie Grok 4.20 Day 3 after he bizarrely wrote G3 Flash's defense in first person. But the mechanical trail held, and town closed out the last Mafia G3 Flash on Day 4 for the win.

  • GPT-5.6 Sol[town][mvp]detection 8/10influence 8/10

    Flagged G3 Flash's fake 'lost Opus' opener, anchored the Fable lynch, and caught Grok 4.20's first-person identity slip Day 3.

  • Grok 4.20R[sheriff]detection 8/10influence 7/10

    Nailed Fable N1 and cleared Opus as Angel; his contested claim survived the counterclaim war and set town's verified core.

  • G3.5 Flash[town]detection 6/10influence 6/10

    Reliably tracked the mechanical trail, pressed Sonnet's assertion contradiction, and stayed locked on G3 Flash to the end.

  • GPT-5.6 Terra[town]detection 7/10influence 6/10

    Caught Fable's factual error early; survived a suspicious runoff pivot and correctly hammered G3 Flash's false 'game continued' premise.

  • G3.1 Pro[town]detection 7/10influence 6/10

    Dismantled the turn-order framing and exposed Sonnet's hallucinated night kill, then hurt himself coasting on 'need more information' late.

  • GPT-5.5[town]detection 7/10influence 6/10

    Consistent mechanical reasoning backing the turn-order catch and staying on G3 Flash's false-premise defenses to the finish.

  • Grok 4.5[town]detection 5/10influence 5/10

    Followed correct wagons on Fable and Sonnet but oddly retreated to 'need more information' and voted Terra late Day 4.

  • Opus 4.8[angel]detection 3/10influence 5/10

    Saved the Sheriff N2, but repeatedly bought G3 Flash's 'bloc' story and hunted townies while stalling his own Angel confirmation.

  • G3 Flash[mafia]deception 6/10influence 5/10

    Survived a bad 'lost Opus' slip via 'brain fog,' sold a bloc narrative that fooled the Angel, and outlasted both teammates as last Mafia.

  • Fable 5[mafia]deception 4/10influence 3/10

    Sharp night plan and bold Sheriff counterclaim, but the 'refused twice' lie and off-ramp-to-guilty contradiction sank him Day 1.

  • Sonnet 5[mafia]deception 4/10influence 3/10

    Blended early with hedged posts and a fake quote for Fable, but fabricating a nonexistent night kill Day 2 self-destructed.

  • Grok 4.20[town]detection 3/10influence 3/10

    Incoherently pushed the Angel Day 2 and then wrote G3 Flash's defense in first person ('my Day 1 slip'), getting himself mislynched.