Mechanisms

The Commitment-Device Gap

The distance between a stated rule and an enforceable one

A commitment device is whatever makes breaking a promise costly. Most systems state intentions and never bind them — a policy in a document, a rule in a prompt, a guideline in a README. The commitment-device gap is the distance between the rule as stated and the rule as enforced, and the incentive lives entirely on the enforced side. If defecting costs nothing, the stated rule is a preference, not a constraint.

Written rules are cheap talk in the technical sense: costless to emit, and therefore carrying no information about what the emitter will actually do. A costly signal is different — it is expensive enough that only someone who means it would send it. The design question for any rule is which of the two you have built.

In agent systems this shows up as instructions that describe a constraint the runtime never enforces. The instruction is real, the intention is real, and nothing in the execution path can refuse the prohibited action. The gap is invisible precisely because the stated rule reads exactly like an enforced one.

Closing it means moving the rule to where the action happens: a check that refuses at the point of the action rather than reporting on it afterwards. A guard that discovers a violation after the fact is a detector. A guard that makes the violation fail is a commitment device.

Where this is worked

• Issue 5: Why Threats Work — The Neural Economics of Cooperation
• Issue 11: The Guard Nothing Reaches

The tool

There is no tool for this one yet. A commitment-device linter is a named candidate in the studio's tool backlog, not a shipped product — it is listed here because the mechanism is real and worked, not because something is buildable today.