1st place (non-manager) — Nightfalls Premier League, Israel's most prestigious league.
10+ years in high-tech.
Profitable poker player.
Before we begin — a change of mindset
“He gets on base.”
Moneyball (2011). A room full of experts argues from gut feel — and keeps losing. One number changes how the game is played. Tonight, we do the same thing to Mafia.
Two days ago · World Cup Final · MetLife Stadium
Spain 1–0 Argentina. Method beat magic.
Possession
ESP
65%
ARG
35%
Shots
ESP
20
ARG
2
Shots on target
ESP
12
ARG
0
Expected goals (xG)
ESP
2.29
ARG
0.22
0
Argentina shots in the first 90 minutes — a first in a World Cup final. Spain's keeper didn't make a single save all night.
774
Passes completed by Spain — a World Cup final record. Half of Argentina's possessions never crossed midfield.
Argentina brought Messi, momentum and magic. Spain brought structure — and optimized the percentages on every single possession. That's the mindset we're borrowing tonight. (Team stats: ESPN, 120 min.)
What this talk covers — and how it works
In this presentation, we'll cover:
How to think about the game — the mindset behind good decisions.
Where wins actually come from — the mechanics that decide games.
How to improve in the long run — not just in the next game.
Basic worked examples that put the principles into practice.
Ground rules
This session is interactive. When I pose a question, jump in — and there are dedicated stops for open questions. Between them, we'll keep the pace and stay on track.
Most of this is consensus, not my opinion. Where it isn't, suspend judgment for now and go with it.
It will click at the end. We start with seemingly disconnected pieces; by the close, they come together into one picture.
1.1 Judge yourself by the long run
The right move is the one that wins more often.
You can play by feel — or you can improve. Improving means optimising a single number: your win rate over many games.
At every decision, choose what wins more games on average — not what feels safe or satisfying.
This is probabilistic thinking. A correct decision still loses sometimes.
Read the situation right and the same call pays off 7, 8 times in 10. The losses are the cost of being correct.
1.2 Read others by self-interest
The core question for every player: did they act in their own best interest as a citizen?
The caveat: weaker players do this less reliably — so a bad-but-innocent player can look guilty.
For you, the asymmetry pays: strong, self-interested play may draw suspicion, but it still wins more games.
Why it works as a signal
A citizen's incentives are fixed and public — everyone knows what's good for the town.
Deviating from your own interest is information. It's a thread to pull.
Part II Where Wins Come From
The engine of the game.
Strip the game to its mechanics, find the mafia's structural edge — then find the town's one answer to it.
2.1 The baseline — play with no logic
If the town used no logic, it loses almost every game.
Imagine citizens voting at random while the mafia simply do what they already can: recognise each other and vote together.
Random town vs. coordinated mafia.
Coordinated black arrows reliably land on red; scattered red arrows rarely land on black. The town's only way to beat this baseline is to add logic.
2.2 The asymmetry — and the tell it creates
The mafia knows. The citizen must infer. That gap leaks.
MAFIA
Sees the whole board. Must pretend not to — fake confusion, fake reads, fake doubt.
Effort to look normal
CITIZEN
Sees nothing. Simply reasons honestly from what's said and done.
Effort to look normal
Because a mafia knows more and must hide it, he behaves differently. Citizens who cross-reference reads force him to play 5× harder just to look ordinary — and that strain is detectable.
2.3 The town's only weapon
Collective logic — anchored on the sheriff's checks.
No single citizen beats the mafia. The group's pooled reasoning does — and it's better than average precisely because it's built around the game's hard information: the sheriff's checks.
⚠ The failure mode: "Superman"
Pool opinions. Make moves that let citizens prove their colour to one another.
Go solo — do whatever you want — and you sabotage your own side.
Red players who "Superman it" hand games to the mafia.
2.4 Logic, applied — a worked example
5 left · 2 mafia · one confirmed red. How do you vote?
A citizen the sheriff has checked red proposes a target. Whatever your own read — you must vote with the confirmed red.
CAN WIN
Vote with the confirmed red. If he's right, the town survives. He's your only path that has any winning world in it.
ALWAYS LOSE
Vote against him. If he's wrong, the two mafia were voting with him anyway — you lose. If he's right, you just saved a mafia. Either way: you lose.
Basic — but it proves such forced moves exist. Accept that, and you can start finding the higher-level moves that win even more.
Part III The Anatomy of a Game
How games are actually won.
The mafia's two routes, what the data says about them, and the phase structure that governs every decision.
3.1 How the mafia wins — two routes
Win the version, or kill the information.
High risk
⚔️
Win the version
Counter-reveal as sheriff and beat the real one head-to-head by being more believable.
Preferred
🌫️
Hide
Destroy the checks and the real information until everything is just "he said, she said."
The town's reasoning is built on the sheriff's checks. So the mafia must either end those checks — or out-argue them outright.
3.2 What the data says
No sheriff, no town. So the mafia should hide.
Sheriff alive & trusted
town wins
Sheriff dead early
mafia wins
Illustrative magnitudes. The lesson is the direction: kill the sheriff before his checks land and the town collapses. So a disciplined mafia ends the check mechanism rather than gambling on a head-to-head — where, at high levels, the town usually picks correctly.
3.3 The spine — every game has three phases
Read the whole game through the sheriff's status.
Phase 1
Sheriff alive & unknown
Mafia: play linear — vote red, work with your team.
Town: gather info, keep the sheriff hidden.
→
Phase 2
Sheriff alive & known
Mafia: kill him — unless too much info is already out.
Town: cash the checks before he dies.
→
Phase 3
Sheriff dead
Mafia: no info left — get tricky, just keep one alive.
Town: survive on inference alone.
This phase is the single most useful lens in the game — it tells both sides how to play right now.
3.4 Rounds & the clock
Round 0 is mostly blanks — maybe one slip, almost no real information.
The 9-player round (after the first night) is where the game truly begins.
From here the town lives in "critical" — one wrong elimination can lose the game outright.
Mafia's ideal: vote out a red and shoot the sheriff. Do both and they almost always win.
Why a live sheriff saves the town
Even in critical, a trusted sheriff guarantees ≥1 correct vote per round.
Via a black check, a fake sheriff exposed against him, or 2–3 red checks to choose from.
And if killed after revealing, he still leaves another check behind.
3.5 The sheriff's dilemma — stay hidden
Reveal in 9, and you trade the engine for a single check.
Reveal in 9 players one check, then dead
✓
→
☠
→
✓
Stay hidden check after check
✓
→
✓
→
✓
In disciplined circles, revealing in 9 without a great reason reads as black. This is also why you never let a "red-check" claimant force a 3-way split — the mafia fakes that every game to drag the real sheriff into the open.
3.6 Information buys eliminations
More checks, more margin for error.
Blind game
Round 0 no vote, one dies at night. You get ~4 eliminations to find 3 mafia.
✓
✓
✓
✗
Wrong even once → you're in critical.
Full-information ceiling
With every check known, the town can split and force its way to ~7 eliminations.
1
2
3
4
5
6
+1
A spare elimination — two versions, both winnable.
An extreme that never literally happens — it marks one end of the scale. The takeaway: when the mafia misses and you hold checks, lean hard into eliminating from a known version.
Part IV Playbooks by Role
Four seats, four playbooks.
Each role's strategy is just the engine, seen from that seat.
4.1 The Citizen
CITIZEN
Eliminate mafia and the most-suspected players.
"Redden" yourself — actively prove your colour.
Extract information; make the mafia sustain a story as long as possible.
Reward the better play. Assume self-interest. Lose some, win more.
Your whole job: feed the engine, then strike late.
Every citizen action either adds information or spends it. Add as much as you can before you're forced to spend.
Seat 5: the confident smile of a "definitely-not-mafia" citizen. 😇
4.1 · nuance When you decide is itself a decision
Decide critical things late. Lock certain things early.
◀ Decide earlyDecide late ▶
Power to win the version you chose
50%
Chance you chose the correct version
50%
Drag the slider — every gain on one side is a loss on the other.
4.2 The Mafia
Be indistinguishable from a citizen — and default to linear.
Play with your team, vote red, behave naturally. Switch to tricky, against-each-other play only once the information is gone (Phase 3, or a doomed Phase 2).
Just two guys. Definitely not the same team. 🕶️
MAFIA
Act unpredictably; never look coordinated.
Know your win condition: hide vs. head-to-head.
Sheriff killable without all mafia exposed → hide.
Otherwise → don't fear the linear attacking game.
Common error: turning on a teammate who's already suspected, or faking fights and getting checked anyway.
For the town, this is a gift: actions before the sheriff is found are far more incriminating than actions after — but a vote that seals a 2-on-2 mafia win is hard information at any time.
4.3 The Sheriff
🛡 SHERIFF
Gather maximum information before revealing; survive, stay hidden.
Act distinguishably from mafia — hide too well and you play exactly as a mafia would.
Make checks that produce as many colours as possible.
Weigh every check by its expected value.
You are the engine. Protect it, and spend each check wisely.
value(check) = P(black) × gain if black + P(red) × gain if red
The sheriff, staying hidden in 9 players. 🤫
4.4 The Don
The orchestrator. Find the sheriff and end the engine.
The Don runs the mafia's game. The whole plan bends toward one outcome: kill the sheriff early, before his checks can pile up.
♛ DON
#1 priority: eliminate a red and kill the sheriff within ≤2 checks — the night after the 9-player round.
Know the win condition: hide vs. reveal-as-sheriff and attack linearly.
Play indistinguishable from a real sheriff.
When killing, prefer a player who'll never be voted out over one with a useful opinion.
Part V Putting It Together
A worked example.
One common situation — and how the engine, the phases, and the long-run math decide it for you.
Example 1 The two-sheriff problem
Two players claim sheriff. Whose version do you play?
1
2
3🛡
4
5
6
7📜
8
9
DEAD Player 7
Voted out in the 9-player round. In his will, claims sheriff — with a black check on the player killed night 1.
vs
ALIVE Player 3
The next day, reveals as the real sheriff — contradicting 7's version.
Play 7's version, 3's version, or both?
Think in win-rate, not in who feels honest.
Play Player 3's version.
The alive, uncontested sheriff is the credible one — and, as the next slide shows, it's the only choice that actually wins games.
Example 1 · why It's the math, not the manners
value = P(version is correct) × P(you win it if you commit)
Player 7's version dead claim
P(7 really is sheriff)plausible
P(win even if he is)low
Expected value
low
Player 3's version alive sheriff
P(3 really is sheriff)strong
P(win if committed)high
Expected value
high
The decisive point: even if you back 7 and he truly is the sheriff, you still lose most of those games. Back 3 and be right, and you win a lot of them. Splitting across both just wastes votes.
The meta point: rewarding a dead claim lets the mafia "declare a different version" for free, every game — the equivalent of letting a revealed sheriff break at 10. Play 3 consistently and you teach both sheriffs and mafia that only that line is rewarded — raising town wins over the long run.
★ If you remember nothing else
The essentials.
01
Play the long run.
Choose the move that wins more games on average. Being wrong sometimes is correct.
02
Logic is the town's only edge.
Random play loses ~99%. Cooperate, pool reads, prove your colour — never Superman it.
03
The sheriff is the engine.
Town: keep him alive, hidden, trusted. Mafia: end the checks before gambling on a reveal.
04
Read the phase, then act.
Sheriff unknown → known → dead tells both sides exactly how to play right now.
In closing
Don't memorise the game. Learn to think it.
You won't leave here knowing every mechanic, split, and situation — those come with games, and time. The point of this talk is the layer underneath: how to reason about the game, so that every game you play makes you better over the long run.
