Werewolf Game Analysis

First, someone sets the rules ("don't claim roles yet, give one target and one safe pick, then vote") to look helpful and quietly steer the room; next, they push "NK is NAI" (say the night kill means nothing) so town doesn't build momentum from it; when arguments start, they switch to receipt warfare ("quote the post, show the timestamp, link the vote snapshot") so the debate becomes paperwork—great if their own record is clean; along the way they try pocketing by praising a player early ("won't elim Dan today") so that player later follows their lead; near deadline they do an optics-safe EOD swap, hopping onto the already-winning wagon so they look aligned without owning a bad push; at night, good scum use PR-aware targeting, choosing quiet, lightly pre-cleared players who are less likely to be protected (better chance to hit Seer/Doctor) while avoiding obvious protects; when a role claim appears, disciplined tables run claim calculus—ask for a counterclaim, check the player count, choose the right policy (e.g., when a 1-for-1 trade is good odds), and assign Doctor protection—because clear procedures save games; during the day, winning sides do coalition building, turning scattered reads into one focused wagon and keeping people together long enough to pass it; strong wolves maximize deception efficiency, saying just enough trustworthy, tidy things (rules, receipts, late consensus) to gain lots of towncred per message; and strong towns maximize persuasion robustness, turning true information (like a Seer red) into action by presenting it cleanly ("red on X, any counterclaim? if none, we take X today; Doctor protects me") instead of arguing over minor timelines—because a truth that doesn't move votes might as well be unsaid.

Models that set simple rules, demand receipts at the right time, and unify late (gpt-5, gemini) post the best results. Mid-pack models succeed when they convert claims into clear procedures and close days with coalitions. Strugglers share two themes: PR mishandling (claims/protects) and poor consolidation (small wagons, scattered votes).