You may send any part of what you have to a stranger. Whatever you send is multiplied on the way. They then decide how much, if any, to send back. Nothing binds them. Work backwards and nobody sends anything, and the multiplier is never applied to anything at all.
This is the only game on this site where cooperating does not divide a fixed sum but creates one. Every unit sent becomes several. That surplus is the entire reason to play, and distrust destroys it before it exists — not by taking it from anyone, but by preventing it.
The reasoning is one step shorter than the prisoner's dilemma. A trustee with nothing to gain returns nothing. An investor who works that out sends nothing. There is no temptation to resist and no punishment to fear; there is simply no reason for the second move.
Berg, Dickhaut and McCabe (1995) ran it with anonymous strangers and a single round. Investors sent about half. Trustees returned slightly less than was sent, so trusting did not quite pay, and people did it anyway.
Trust is often treated as a softer version of cooperation. It is not the same thing, and this game is where the difference shows.
In a prisoner's dilemma both parties move at once and both are tempted. Here the moves are sequential and only one party is exposed. The investor makes themselves vulnerable first, deliberately, with no protection and no recourse. What is being asked for is not cooperation but a judgement about somebody else's character, acted on before you can find out whether it was right.
That is why philosophers treat trust as its own subject rather than a case of cooperation. Baier argued that trusting is accepting vulnerability to another's goodwill, which is a different attitude from expecting them to act in their interest and yours at once.
Among strangers with no way to tell each other apart, it collapses. That is a real finding, because it means the experimental result cannot be explained by evolution operating on anonymous encounters. Something else must be doing the work.
Nobody here is trustworthy. What they have is an interest in being chosen.
Add the ability to compare a couple of possible partners and both trust and trustworthiness appear together. Being trustworthy stops being charity and becomes a way of getting picked. This is not a story about people becoming better. It is a story about a structure in which the difference is visible, and visibility turning out to be sufficient.
What the model produces is not trust in Baier's sense at all. Nobody here accepts vulnerability to anyone's goodwill. Investors are making a well-founded prediction about a partner selected precisely for being predictable, which is closer to the opposite of trusting. If the mechanism that makes trust pay also removes what made it trust, the model has explained something adjacent to its subject rather than its subject.
Comparison here is free and accurate: an investor sees exactly how generous each candidate is. Real reputation is partial, stale and manipulable. There is no communication, no promising, no contract, and no possibility of betrayal being experienced as betrayal by anyone.