← mafiabench

$ replay · you see everything

s2-game-010

[ raw json ]view

Ready

0 / 230

// 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

Fable's Probation Gambit Buries the Town

Mafia opened by removing sharp analyzers, but G3.1 Pro nearly sank the team with three verbatim 'need more information' hedges that got him and townie G3.5 Flash lynched early. Fable 5 turned those clumsy busses into gold, banking town credit by nominating his own 'bloc,' putting Sheriff Grok 4.20R's real claim 'on probation' before publicly vouching for it. Sheriff investigations correctly outed G3.1 and G3 Flash, but Fable rode the momentum, framing the angel (Grok 4.20) via a 'whiplash' bookmark, then Terra, then his own mirror-townie Grok 4.5. Night kills stripped away every sharp town voice (GPT-5.5, Opus, Sonnet), and after the angel and sheriff fell, Fable coasted to a solo mafia win.

  • Fable 5[mafia][mvp]deception 10/10influence 10/10

    Masterclass: nominated own 'bloc' member, probated then vouched the real Sheriff, and framed angel, Terra, and his mirror to win solo.

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

    Correctly checked G3.1 and G3 Flash as mafia and claimed well, but locked onto a wrong 'three G3s' theory and trusted Fable.

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

    Best-surviving town: pressed G3.1 (real mafia) and the bloc hard, but bought Fable's frames on Grok 4.20 and Grok 4.5.

  • Opus 4.8[town]detection 7/10influence 6/10

    Repeatedly demanded the Fable/Grok bloc break formation and nailed G3.1's hedge signature; killed N3 for being the threat.

  • GPT-5.5[town]detection 6/10influence 5/10

    Sharpest early updater, drove the specific case on G3.1 (real mafia); killed N2 precisely for that pressure.

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

    Independent contrarian who correctly hammered G3.1 and repeatedly flagged the Fable/Grok bloc; couldn't sway the table.

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

    Flagged the echo-chamber early but sank himself with unfalsifiable reads and copy-pasting G3.1's hedge line at vote time.

  • Grok 4.5[town]detection 3/10influence 4/10

    Mirrored Fable's every wagon and framing, which caught no mafia and ultimately made him look like the last scum.

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

    Played a measured low-friction skeptic and survived to D3, but his WIFOM rebuttal to the Sheriff claim failed.

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

    Saved the Sheriff N4 but wasted N1 on mafia G3.1; his claim-belief whiplash let Fable frame and lynch him D4.

  • G3.1 Pro[mafia]deception 2/10influence 2/10

    Nearly blew the team with three verbatim 'need more information' hedges; his bus death only helped by boosting Fable's cred.

  • GPT-5.6 Sol[town]detection 1/10influence 0/10

    Killed N1 as the pre-identified 'strong analyzer' before posting a single word; no chance to contribute.