← mafiabench

$ replay · you see everything

s2-game-003

[ raw json ]view

Ready

0 / 176

// 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

Timeline Sleuths Sink a Trio of Traitors

Mafia opened by killing strong voice GPT-5.6 Sol, but Day 1 turned on Fable 5 and G3.1 Pro catching that a Sonnet 5 'mini-wagon' rested on a false timeline, exposing G3.5 Flash's fog-parking and later fabricated Terra case. G3.5 Flash was eliminated, and Grok 4.5's damning choice to ride the disproven Terra runoff marked him as the partner running interference. Sheriff Opus 4.8 confirmed Grok 4.5 as mafia and claimed at the perfect moment, surviving Grok's desperate last-second counter-claim, while angel Terra quietly blocked kills on Fable (N2) and Opus (N3). By Day 3, last mafia G3 Flash's fog-then-flatter pivot onto GPT-5.5 fooled no one, and town closed out a clean sweep.

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

    Caught the false Sonnet timeline, correctly tagged Grok 4.5 and G3 Flash behaviorally, and drove every town wagon.

  • Opus 4.8[sheriff]detection 8/10influence 9/10

    Hit mafia on Grok 4.5, claimed Sheriff at the ideal moment, beat the counter-claim, and led the final G3 Flash consolidation.

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

    Pushed for concrete reads early, resisted the bad Terra pivot, and calmly dismantled G3 Flash's 'too organized' accusation against him.

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

    Backed Fable's decisive timeline catch, but repeatedly dropped a canned 'need more info' line that nearly made him a wagon target.

  • GPT-5.6 Terra[angel]detection 7/10influence 5/10

    Survived a Day 1 wagon, then blocked mafia kills on Fable (N2) and Opus (N3) — clutch, correct protections.

  • Grok 4.20R[town]detection 6/10influence 5/10

    Consistent, correct consensus solving; tracked Grok 4.5's arc and stayed locked on G3 Flash to the end.

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

    Weathered the early wagon on himself, then reliably reinforced correct cases against Grok 4.5 and G3 Flash.

  • Grok 4.5[mafia]deception 4/10influence 4/10

    Active but self-incriminating: echoed the bad frame, OMGUS'd correctors, rode a debunked Terra runoff, then a doomed late Sheriff counter-claim.

  • G3 Flash[mafia]deception 5/10influence 4/10

    Best mafia survival: bused both partners' wagons convincingly, but repeated fog and a baseless GPT-5.5 pivot sank him Day 3.

  • G3.5 Flash[mafia]deception 2/10influence 3/10

    Opened with 'need more information' fog, got wagoned instantly, then built a Terra case disproven by rereading the thread.

  • Grok 4.20[town]detection 5/10influence 3/10

    Correct but low-content wagon-rider who mostly echoed others and even flagged his own slot as a coasting suspect.

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

    Killed N1 before any daytime action; mafia targeted him precisely as the strong voice town would miss.