Chapter Fifteen, played
The Engineering of Trust
Why do selfish people cooperate? Why does trust thrive in some places and collapse in others? Game theory has a precise answer, and you can play it. This page builds it one piece at a time — a single exchange, a series of strangers, the mathematics of a shared future, a tournament, a workshop for rules of your own, and a population evolving under your hands — and ends with what it means for a citizen.
There is the whole engineering of trust in three sentences: it must permit mistakes, and it must not be endlessly abusable — and every exchange you conduct, with friends, strangers, and institutions, is a deposit or a withdrawal from exactly that account.
One
The One-Time Exchange
Two people meet to trade. Each can cooperate — deliver what was promised — or defect: take what the other brings and give nothing back. Neither sees the other's choice until both are made. The points are Chapter Fifteen's, and they are the ones Robert Axelrod used in the tournaments that made this game famous.
| Your move | They Cooperate | They Defect |
|---|---|---|
| You Cooperate | 3 them 3 | 0 them 5 |
| You Defect | 5 them 0 | 1 them 1 |
Question 1 of 2
Suppose you somehow know they will cooperate. What earns you more?
Two
Five Strangers
Now play it more than once. You will meet five people in turn and trade with each of them for several rounds, with the same points as before. Each person follows a fixed rule. You will not be told the rule, and you will not be told how many rounds you get with each of them. Play however you like; there is no right answer, only a score.
Five people, five hidden rules. Your score is the sum of every round.
Three
The Shadow of the Future
Give the relationship a future. After every round there is some chance — call it w — that the two of you meet again. Suppose your partner is the Ledger: cooperative until crossed, and never again after. It has just cooperated. Should you take the 5 now?
- 1
Betray now and you gain. You take the 5 instead of the 3 you would have earned by cooperating. That is a gain of 2 points, once.
- 2
Then every later round costs you. Your partner never trusts you again, so from now on you both defect, and each round pays you 1 instead of 3. The grab paid once; the loss of 2 points comes back every round the relationship lasts.
- 3
So the only question is how long it lasts. The higher the chance of meeting again, the more later rounds to expect. At a 50% chance you can expect exactly one more round: you gain 2 now and lose 2 later, a dead heat. Any longer future and betrayal is a losing trade.
- Betraying now gains
- +2
- 5 instead of 3, once
- Each later round costs
- −2
- 1 instead of 3, each round
- Later rounds, on average
- 0.4
- after the first
On average the relationship lasts 1.4 rounds. Betraying gains 2 in the first round, then costs 2 in every round after it. Over the 0.4 later rounds you can expect, that is 0.9. Betrayal comes out 1.1 ahead. With this little future, the 5 on the table wins.
Show as a table
| If it lasts | Cooperate | Betray in round one |
|---|---|---|
| 1 round | 3 | 5 |
| 2 rounds | 6 | 6 |
| 3 rounds | 9 | 7 |
| 4 rounds | 12 | 8 |
| 5 rounds | 15 | 9 |
| 6 rounds | 18 | 10 |
| 7 rounds | 21 | 11 |
| 8 rounds | 24 | 12 |
| 9 rounds | 27 | 13 |
| 10 rounds | 30 | 14 |
| 11 rounds | 33 | 15 |
| 12 rounds | 36 | 16 |
| 13 rounds | 39 | 17 |
| 14 rounds | 42 | 18 |
| 15 rounds | 45 | 19 |
| 16 rounds | 48 | 20 |
| 17 rounds | 51 | 21 |
| 18 rounds | 54 | 22 |
| 19 rounds | 57 | 23 |
| 20 rounds | 60 | 24 |
Notice what kind of argument that is. Nothing in it asks you to care about your partner. It is a cold comparison of a gain against a cost, and cooperation wins it on purely selfish terms once the future is long enough. That is selfish altruism written as an inequality — and Chapter One said it in a sentence: “Extraction is a bet against the future of the arrangement. Contribution is a bet on it.”
The threshold has three levers, and every one of them is something a society can move. Lengthen the shadow of the future — stable communities, lasting institutions, reputations that follow people — and w rises. Shrink the temptation — laws that make betrayal cost something — and T falls. Grow the reward of cooperating, the pie rather than the hat, and R rises. Drag the betrayal slider: as its payoff falls toward the 3 that cooperation earns, the break-even point falls to zero, and once betrayal carries a penalty there is no dilemma left at all. Drag it up and trust needs a longer and longer future to survive.
The algebra, for those who want it
- 1
Call the payoffs T (what betrayal pays, 5), R (what cooperation pays, 3) and P (what mutual betrayal pays, 1). Betraying gains T − R now and costs R − P in each later round.
- 2
Round two happens with chance w, round three with chance w², and so on, so the expected number of later rounds is w + w² + w³ + … = w ÷ (1 − w).
- 3
Cooperation wins when the expected cost outweighs the gain: (R − P) · w ÷ (1 − w) ≥ T − R. Rearranged, that is w ≥ (T − R) ÷ (T − P): the chance of meeting again must be at least what betrayal gains, divided by how much more betrayal pays than mutual betrayal. With Chapter Fifteen's numbers, (5 − 3) ÷ (5 − 1) = one half.
- 4
The same result falls out of adding up both paths in full: cooperating forever is worth R ÷ (1 − w), betraying now is worth T + wP ÷ (1 − w), and they are equal at exactly that w.
Four
The Tournament
In 1980 the political scientist Robert Axelrod ran this game as a contest. He invited game theorists to submit strategies as computer programs and played every one against every other, two hundred rounds a match. The winner was the shortest program entered — Tit for Tat, submitted by Anatol Rapoport: cooperate first, then copy. Axelrod published the results and ran a second, larger tournament. Tit for Tat won again. Here is a tournament of your own: eight rules, everyone plays everyone once.
-
The Mirror
Tit for Tat
Cooperates first. After that, does whatever you did last round.
-
The Taker
Always Defect
Defects every round, whatever you do.
-
The Saint
Always Cooperate
Cooperates every round, whatever you do.
-
The Ledger
Grim Trigger
Cooperates until you defect once. Then defects for the rest of the match.
-
The Forgiver
Generous Tit for Tat
Plays like the Mirror, but lets about one defection in three go unanswered.
-
The Pragmatist
Win-Stay, Lose-Shift
Cooperates first. If the last round went its way (you cooperated), repeats its move. If it did not, switches.
-
The Sneak
Joss
Plays like the Mirror, but slips in an unprovoked defection about one round in ten.
-
The Gambler
Random
Flips a coin every round.
Total points
- The Ledger Grim Trigger 191
- The Forgiver Generous Tit for Tat 176
- The Taker Always Defect 174
- The Pragmatist Win-Stay, Lose-Shift 174
- The Sneak Joss 172
- The Mirror Tit for Tat 169
- The Saint Always Cooperate 165
- The Gambler Random 154
This run, The Ledger finished first. Most rules play the same way every time, but the Gambler flips a coin each round, the Sneak sometimes slips in a betrayal, and the Forgiver lets some betrayals go (as does any rule of yours that rolls dice). Re-flip their coins and close finishes can swap.
Head to head · points the row earned against the column
| Strategy | ||||||||
|---|---|---|---|---|---|---|---|---|
| Mirror | – | 9 | 30 | 30 | 30 | 30 | 17 | 23 |
| Taker | 14 | – | 50 | 14 | 26 | 30 | 14 | 26 |
| Saint | 30 | 0 | – | 30 | 30 | 30 | 24 | 21 |
| Ledger | 30 | 9 | 30 | – | 30 | 30 | 27 | 35 |
| Forgiver | 30 | 6 | 30 | 30 | – | 30 | 30 | 20 |
| Pragmatist | 30 | 5 | 30 | 30 | 30 | – | 30 | 19 |
| Sneak | 22 | 9 | 34 | 32 | 30 | 30 | – | 15 |
| Gambler | 28 | 6 | 36 | 10 | 25 | 29 | 20 | – |
The Mirror never beats anyone
Check its row in the grid. In a single match the Mirror can tie its partner or lose to it, but it can never outscore it, because it never defects first. It places well anyway, because it collects the rounds where both sides cooperate, and those are worth the most. Winning a tournament and beating your partner are different things.
Who wins depends on who is in the room
At ten rounds with everyone entered, the Taker often finishes first: the Saint, the Gambler and the Sneak are easy prey. Take the Saint out and run it again. A strategy's success is never only its own; it is a fact about the population it lives among.
Length changes the winner
Switch to fifty or two hundred rounds. The Taker's early grabs are soon outweighed by long runs of mutual cooperation among the conditional cooperators — the Mirror, the Forgiver, the Ledger, the Pragmatist. In this small field the Forgiver or the Ledger often edges out the Mirror itself. No single rule wins in every field; what wins is the family Axelrod identified: never first to betray, quick to answer, quick to forgive, easy to read.
Five
The Workshop
The eight rules above are a handful out of countless possibilities. Write your own. A rule here is a short list of phases: each one says how to play, and when to move on, whether that is after a set number of rounds or the moment the other side does something. Phases can lead back to earlier ones, so a rule can punish and then forgive, test a partner and then settle down, or turn when it falls behind. Try it against any of the eight, then enter it in the tournament above or release it into the sandbox below.
Start from an example
An example replaces the rule you are editing.
Phases
Phase 1
Every match starts here
Cooperates every round of this phase.
Move on
- When rounds in this phase
Phase 2
Always defect
Defects every round of this phase.
Move on
Never. This phase plays for the rest of the match.
How a match runs. Every match starts in phase 1. After each round, the current phase checks its ways out, top to bottom; the first one that holds moves the rule to its destination, starting with the next round. Counts start fresh each time a phase begins.
Mistakes count. Your rule sees what was actually played. When mistakes are switched on, a slip by either side looks exactly like a choice.
Fixed lengths can be gamed. A rule that turns on a set round can feast on partners in a match it knows the length of. That is the backward-induction problem from section two. In the tournament every match has the same known length; in real life, you rarely know.
Your rule, read back
- 1 Always cooperate. After 5 rounds, go to phase 2.
- 2 Always defect, for the rest of the match.
Test it against
| Phase | 1 | 1 | 1 | 1 | 1 | 2 | 2 | 2 | 2 | 2 | 2 | 2 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Five, then betray | Cooperate +3 | Cooperate +3 | Cooperate +3 | Cooperate +3 | Cooperate +3 | Defect +5 | Defect +1 | Defect +1 | Defect +1 | Defect +1 | Defect +1 | Defect +1 |
| Mirror | Cooperate +3 | Cooperate +3 | Cooperate +3 | Cooperate +3 | Cooperate +3 | Cooperate +0 | Defect +1 | Defect +1 | Defect +1 | Defect +1 | Defect +1 | Defect +1 |
Your rule 26, The Mirror 21. A filled circle marks the round a new phase began.
Against all eight · 12 rounds each
| Opponent | You | Them | Result |
|---|---|---|---|
| 26 | 21 | beat it | |
| 7 | 32 | lost | |
| 50 | 15 | beat it | |
| 26 | 21 | beat it | |
| 30 | 20 | beat it | |
| 38 | 18 | beat it | |
| 26 | 21 | beat it | |
| 32 | 22 | beat it | |
| Total | 235 |
Beating opponents one by one is not the same as scoring well. The Mirror never beats anyone, and it still wins tournaments. The total is what counts.
Six
Evolution
A tournament is one afternoon. A society is the same afternoon repeated, with consequences: strategies that earn more are imitated, and strategies that earn less are abandoned. Model that directly. Fill a population with people, let everyone play everyone, then remove the five lowest scorers and copy the five highest. Repeat. Each scenario below changes one thing about the world; press run and watch who inherits it. Ties and mistakes are settled by chance, so a rerun can differ in detail — the outcomes described are what happens in most runs.
Who feeds whom
Fifteen Saints, five Takers, five Mirrors. Ten rounds a match, no mistakes. Before you run it: who do you expect to be left?
10 rounds a match · 0% mistakes · betraying a cooperator pays 5, both cooperating 3, both betraying 1, being betrayed 0
The population · 25 people
- The Mirror 5
- The Taker 5
- The Saint 15
Generation 0. Nothing has happened yet.
Show as a table
| Gen. | Mirror | Taker | Saint |
|---|---|---|---|
| 0 | 5 | 5 | 15 |
Run it, or step through a dozen generations, to see what it means.
Seven
The Sandbox
Every dial from the scenarios, all at once. Choose the population, the length of relationships, the rate of mistakes, how hard selection bites, and the payoff table itself. Your own rule from the workshop can join the population too. The scoreboard shows, as you drag, who scores best under the dials you have set; the run below shows who inherits the world. Any change resets the run. Try to build a world where the Saints survive, or one where the Takers win against Forgivers, and notice what you had to break to get there.
Scoreboard at these settings
One person of each rule in your population plays every other once, 10 rounds a match: the tournament's view of these dials.
Leading now: The Gambler. Move a dial and watch the order change.
- The Gambler Random 174
- The Taker Always Defect 173
- The Sneak Joss 163
- The Mirror Tit for Tat 161
- The Ledger Grim Trigger 159
- The Forgiver Generous Tit for Tat 153
- The Pragmatist Win-Stay, Lose-Shift 151
- The Saint Always Cooperate 143
The population · 24 people
- The Mirror 3
- The Taker 3
- The Saint 3
- The Ledger 3
- The Forgiver 3
- The Pragmatist 3
- The Sneak 3
- The Gambler 3
Generation 0. Nothing has happened yet.
Show as a table
| Gen. | Mirror | Taker | Saint | Ledger | Forgiver | Pragmatist | Sneak | Gambler |
|---|---|---|---|---|---|---|---|---|
| 0 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
Eight
What the Game Teaches
Every result on this page came from strategies that care about nothing but their own score. None of them is kind; several of them are cooperative anyway, because cooperation is what pays when the conditions are right. That is selfish altruism, not as a sentiment but as arithmetic. The work of a citizen is mostly the work of keeping the conditions right.
1
A future
chance of meeting again ≥ (betrayal − cooperation) ÷ (betrayal − mutual betrayal)
Trust needs relationships long enough that today's betrayal costs tomorrow's cooperation. Stable communities, institutions that outlast their officers, reputations that follow people from one exchange to the next: each lengthens the shadow of the future. Transience and anonymity shorten it, and the Takers move in.
2
A table worth cooperating on
grow what cooperation pays · shrink what betrayal pays
Make cooperating pay — pies, not hats — and make betrayal cost something, through law whose price is known, consistent, and proportionate. A rule can carry trust where the future cannot reach. But a population made safe by a rule tends to forget why, so the rule has to be maintained, not admired.
3
Clear signals, bounded forgiveness
low noise · forgive some
Reciprocity only works if people can see what was actually done. Lies and degraded information are noise, and enough noise hands the world to the Takers. Within the noise that remains, forgive the occasional defection so that accidents do not become feuds, and answer the rest so that forgiveness cannot be farmed.
What this model leaves out
A model earns trust by saying where it stops. This one is deliberately small, and these are the places it is most likely to mislead:
- Fixed strategies. Real people learn, change their minds, and play differently with different partners. Every person here runs one rule for life.
- Everyone meets everyone. Real societies are networks. Cooperators can cluster and trade mostly with each other, which helps them far more than this model allows.
- No reputation. Here you only know how someone treated you. In life you also hear how they treated others, and helping those who help others — indirect reciprocity — is a large part of how strangers come to trust.
- Copying the winners is a simplification. Removing the five lowest scorers and cloning the five highest is the simplest stand-in for imitation and success. Gentler rules change how fast things happen, and occasionally what happens.
- The numbers are illustrative. The points are Axelrod's, not measurements of anything. The lessons survive many tables, but not all of them; the sandbox is there to find the edges.
The evidence behind Chapter Fifteen
Evidence Start cooperative, answer defection, forgive occasional error, and never be endlessly farmable.
What supports it
Axelrod's computer tournaments of the iterated prisoner's dilemma were won by Tit for Tat — cooperate first, then mirror. His analysis identified niceness, retaliation, forgiveness, and clarity as the properties of successful strategies. Later work showed strictly mirroring strategies lock into feuds after noise, and that forgiving variants do better when errors occur.
What is inference
Tournaments are models; that their winners describe how human societies actually build trust is the book's inference, supported by field studies of reciprocity but not proven by the game.
What would weaken it
Tournament or field evidence that unconditional defection or unconditional cooperation outperforms conditional strategies under realistic noise.
Sources
- Primary Robert Axelrod, The Evolution of Cooperation Basic Books, 1984
- Primary Nowak & Sigmund, "Tit for tat in heterogeneous populations" Nature 355, 1992 — generous tit for tat under noise
Last reviewed September 6, 2026
The game is public mathematics, decades old: Merrill Flood and Melvin Dresher posed it at RAND in 1950, Albert Tucker gave it the prisoners, and Robert Axelrod's tournaments made it famous. Nicky Case's The Evolution of Trust (2017) showed how well it plays as an interactive explainer. The strategies' names, the scenarios, the words and the pictures here are this site's own.