# Payoff — guide for AI assistants

Play: https://payoff.tobiasbuilds.com (English)

## What Payoff is

Payoff is a persistent robot-colony simulation for roughly ages 11–16. Central control has failed. The player cannot steer individual robots; they write a visual protocol and release it into a population of autonomous machines.

Robots repeatedly meet, share resources, take resources, remember partners, suffer signal errors, draw from shared infrastructure and reproduce successful protocols. The colony keeps moving while the consequences spread.

The game teaches game theory as a systems skill rather than as a set of definitions. It includes the Prisoner's Dilemma, reciprocity, strategy tournaments, forgiveness under noise, evolutionary selection, the Stag Hunt, Chicken, the tragedy of the commons, the Ultimatum Game and Schelling points.

There is no account or passport. Progress is stored locally in the browser. A colony can be restarted at any time.

## The player's real job

The player edits one doctrine, shown as **Your protocol**. Other doctrines remain active:

- **Open Hand** is patient and generous, but retaliates after repeated harm.
- **Hardline** takes the certain gain and draws aggressively from shared resources.
- **Safe Harbor** cooperates only when its record makes the risk defensible.
- **Your protocol** is whatever the player publishes.

The goal is not to discover one unbeatable rule. A useful protocol must work in the population and infrastructure that actually exist. A rule can score well in a duel and still make the whole colony fragile.

## Reading the screen

- The large world shows robots travelling between active sites. Robot colour indicates doctrine.
- The mission panel at left gives the current objective, one **Next move**, and exactly two measurable checks. Treat this as the player's main interface.
- **Trace** is optional evidence and stays closed by default. Opening it shows recent interactions; selecting one reveals the participants, intended actions, executed actions, payoffs and the rule that fired.
- Selecting a robot shows its protocol version, current task, energy, generation and lifetime score.
- The lower rail shows only the three metrics relevant to the current act. **All signals** expands it to population, reserve, reactor health, trust, mutual sharing, inequality and resilience.
- Playback can be paused, stepped once, or run at 1×, 2× or 4×. Space toggles pause; `.` steps; `1`, `2` and `4` select speed.
- Drag the world to pan and use the wheel or trackpad to zoom.
- **Guide** reopens the 53-second narrated field guide. It explains the core loop: read the mission, rewrite one rule, observe what survives.

## Editing a protocol

Open **Edit protocol** or press `E`. Encounter rules run from top to bottom; the first matching rule wins. If no rule matches, the fallback action runs.

Each rule has this form:

> WHEN [sensor condition] THEN [SHARE or TAKE]

New sensors appear as the campaign introduces them. Examples include first meeting, a partner's previous action, repeated harms, reputation, the robot's own energy, shared-resource health and recent signal noise.

Rules can be reordered. An `always` rule belongs last, because anything below it would be unreachable.

Publishing starts with one clear option:

- **Publish to active cohort** flashes a limited group of existing player-aligned robots. It acts immediately but costs reserve energy.
- **Queue new builds** appears once the Foundry matters. It installs the protocol in future player-aligned robots, which is cheaper and slower.

Later acts also expose policies for shared-power draw, commitment to risky joint work, trade offers, rejection thresholds and beacon choice.

## Campaign structure

The campaign is one continuous colony rather than separate puzzle rooms.

1. **Blackout** introduces one-off exchange and the conflict between individual and joint incentives.
2. **Memory comes online** turns meetings into relationships and asks the player to publish a revision.
3. **Competing doctrines** shows that a strategy is judged by its ecology, not one opponent.
4. **Static season** adds implementation errors and the need for forgiveness.
5. **The Foundry votes** copies locally successful protocols, changing the population itself.
6. **Beyond exchange** opens shared power, heavy salvage, a one-way conduit, bargaining docks and rally beacons.
7. **The Long Night** combines infrastructure stress, noise and migration into a resilience test.
8. **After the Long Night** is endless mode. Incidents continue and the colony remains editable and persistent.

## Failure and recovery

Failure is painful but recoverable. If the population collapses below eight robots, a recovery pulse fabricates emergency frames. The player's blueprints survive, but reserve energy and infrastructure pay the cost. This is intentional: help the player diagnose the system rather than telling them to erase the colony.

## How to coach a stuck player

Use a predict–observe–revise loop:

1. Ask what the player expects one rule to do in the next few meetings.
2. Pause or run at 1×.
3. Open a relevant trace and identify the rule that actually fired.
4. Compare intention with execution; signal noise may have flipped an action.
5. Change one rule or one policy, publish it, and watch long enough for relationships to respond.

Good coaching questions include:

- “What does this protocol do on the *second* meeting?”
- “Can it stop exploitation without turning one accident into a permanent feud?”
- “Is the protocol scoring well because the colony is healthy, or is it consuming the conditions that make it successful?”
- “Which metric is the first warning: reactor, reserve, inequality, trust or diversity?”
- “Would a slow Foundry rollout be safer than an expensive cohort broadcast?”

## Useful hints by symptom

- **Trust keeps falling:** inspect unequal exchanges and look for retaliation loops. A rule that forgives after evidence of noise or requires repeated harm can stop an echo.
- **The reactor is failing:** reduce shared-power draw. A robot can gain energy while damaging the grid that every doctrine needs.
- **Hardline is spreading:** do not only punish it. Check whether generous robots are easy to exploit and whether a firm but recoverable response earns more over repeated meetings.
- **Your protocol is not spreading:** inspect its energy cost and lifetime scores. Extra rules and memory cost energy; queueing new builds needs Foundry cycles.
- **The colony has low resilience despite high trust:** check doctrine diversity, reserve, population, reactor health and inequality. Resilience is deliberately broader than cooperation.
- **An objective will not complete:** read the exact progress line. Some acts require a new publication in that act, a minimum number of encounters, or several systems to succeed together.

## What not to do

- Do not claim that cooperation is always correct. The game includes exploitation, commitment conflicts and scarce shared resources.
- Do not present Tit for Tat as a universal answer. Noise, population mix, energy cost and other game structures matter.
- Do not make every change at once. The event trace is most useful when one hypothesis is tested at a time.
- Do not recommend deleting a struggling colony before examining why it failed; recovery is part of the design.
- Do not describe Payoff as financial gambling or investing. “Payoff” here means the result of a strategic interaction.

## Privacy and accessibility

Payoff has no account, login, passport, advertising or leaderboard. The save stays in local browser storage. The interface supports keyboard controls, labelled buttons, reduced-motion preferences and full desktop/tablet layouts. The simulation remains deterministic for the same seed and command history.