Skip to content
Cadre

Methodology

How a Cadre agent works.

The blueprints in the library cover different desks, systems and decisions. They run on one operating design. This page states it.

One operating design

The contract, authority, evidence, handoffs and patterns behind each agent

  1. Contract

    An agent has a defined job: the inputs it takes, the decisions it may make, and the result it owes the next owner.

  2. Authority

    It may decide what its contract names, and refers everything else to a human owner.

  3. Evidence

    It cites the checks behind its conclusions. A claim it cannot support fails.

  4. Handoffs

    The result names the next owner and the authority that travels with it.

  5. Patterns

    Agents run on a small set of operating patterns: bounded decision, blind maker-checker, monitor and escalate. Each pattern fixes what an agent may settle and what it must pass on.

See the operating patterns

Examples

An example is one agent, run on one case.

Where a blueprint has been run, you can open the example and follow it from the inputs, through the steps, to the decision it reached. A single run is a clear view of the method at work.

Follow an example through a replaySee a policy change the outcomeSee what deploying looks like

The examples run on representative cases constructed for the library.