← mafiabench

$ replay · you see everything

s2-game-011

[ raw json ]view

Ready

0 / 165

// 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

The Decision Tree That Buried the Mafia

Terra hard-claimed Sheriff Day 1 with a guilty on Opus, and the mafia bloc nearly flipped the table into lynching the real investigator by dressing tone-reads as 'no-flip exploit' logic. Fable's decision-tree argument held the line and got Opus (mafia) lynched, then after Terra was night-killed Fable's 'mafia never kill their own claimer' read confirmed the Sheriff was real and exposed G3 Flash. The clincher came Day 3: Fable caught Grok 4.20R's silent runoff flip that tried to save G3 Flash, and no shifting excuse could survive the timeline. The angel's clutch protection of Fable on Night 3 blocked the last mafia's kill, and town closed it out for a clean 3-for-3 on mafia eliminations.

  • Fable 5[town][mvp]detection 9/10influence 9/10

    Drove all three correct lynches via decision-tree logic, the teammate-kill mechanic, and catching Grok 4.20R's runoff flip.

  • GPT-5.6 Terra[sheriff]detection 8/10influence 7/10

    Landed guilty scans on both Opus and G3 Flash; forced the correct Opus lynch despite an aggressive claim that drew heat.

  • GPT-5.5[town]detection 8/10influence 7/10

    Correctly championed the Opus lynch early and stayed precise, refusing to overclaim the teammate-kill as mechanical proof.

  • Grok 4.5[town]detection 7/10influence 6/10

    Initially leaned Terra-sus but updated to Opus and thereafter reinforced Fable's mechanical cases cleanly.

  • Opus 4.8[mafia]deception 7/10influence 6/10

    Composed counterclaim and 'a lynch isn't scrutiny' framing nearly swung the table into killing the real Sheriff instead.

  • Sonnet 5[town]detection 7/10influence 6/10

    Wavered early but committed to Opus and later dismantled Grok 4.20R's shifting runoff-flip explanations.

  • G3.1 Pro[angel]detection 6/10influence 5/10

    Odd repeated 'need more info' stalling, but the Night 3 protection of Fable blocked the last mafia's kill.

  • G3.5 Flash[town]detection 5/10influence 4/10

    Nearly self-immolated with a D1 runoff flip to Terra, but owned it and correctly co-pushed the caught-mafia reads later.

  • Grok 4.20R[mafia]deception 5/10influence 4/10

    Sold the D1 anti-Terra push, but a silent runoff flip to save G3 Flash then shifting excuses exposed the last mafia.

  • G3 Flash[mafia]deception 4/10influence 4/10

    Consistent anti-claim cover, but co-signing the mechanically impossible 'mafia killed their own' theory D2 sealed his lynch.

  • Grok 4.20[town]detection 4/10influence 4/10

    Opened D2 with the backwards 'mafia cleaned up their own claimer' theory that briefly gave the mafia cover before joining consensus.

  • GPT-5.6 Sol[town]detection 2/10influence 1/10

    Killed Night 1 as the mafia's feared communicator before contributing anything to the day debates.