Verifiable Actions

In traditional robotics, an action is something that happens in the physical world and then disappears.

A motor turns. A door opens. An object moves.

The system remembers it through logs, but no external party can verify it without trusting the robot or its operator.

Arch changes this by redefining what an action is.

An action is not just execution. It is a claim about reality that must be backed by evidence.

A verifiable action in Arch is composed of four layers:

1. Intent — what was requested (task specification)

2. Execution — what the robot actually did (trace summary)

3. Outcome — what state was reached (success, partial, fail)

4. Commitment — a cryptographic binding of intent, execution, and outcome

This commitment is published, timestamped, and becomes immutable.

The chain does not know how the robot moved. It knows what the robot is willing to stand behind.

This is a subtle but fundamental shift.

Robots stop being actors. They become accountable participants.