Chains

A governance programme is not a collection of isolated mechanisms. In practice, mechanisms depend on each other: a risk classification mechanism produces a risk register that feeds a conformity assessment mechanism, which produces documentation that feeds a regulatory reporting mechanism. Chains make these dependencies explicit.

What a chain is

A chain is a directed graph of mechanisms connected by their inputs and outputs. Each mechanism in the chain is a node; the edges between nodes represent IO connections — an output from one mechanism designated as an input to another.

Chains are built in Mechanic's chain editor, which visualises the mechanism graph as a flow diagram.

[Screenshot needed] PLACEHOLDER IMAGE: The Mechanic chain editor showing several mechanism nodes connected by directional edges, representing a governance sequence from risk classification through to regulatory reporting

Building a chain

From the Mechanic home page, select New chain.

Name the chain after the governance sequence it represents — "AI Lifecycle Governance", "Incident-to-Remediation", "Risk Classification Pipeline".

In the chain editor, add mechanisms to the chain and draw connections between them by linking an output from one mechanism to an input of another. Connections are only possible where the output type of one mechanism is compatible with the input type of the next.

As you build the chain, the dependency structure becomes visible. You can see at a glance which mechanisms are upstream of others and where the critical path runs.

What chains reveal

The diagnostic value of chains is in the dependencies they surface.

Single points of failure. A mechanism in the middle of a chain that is in Critical health does not just affect its own controls — it breaks every downstream mechanism that depends on its output. Chains make these cascade risks explicit.

Coverage gaps. A chain that starts with a detection mechanism but has no connected notification mechanism shows that detection exists without a defined response pathway.

Proportionality. A chain calibrated for a high-risk AI system might require more rigorous mechanisms at each step than the equivalent chain for a lower-risk system. Scope settings on each mechanism determine which systems each part of the chain applies to.

Chain-level metrics

Chains can carry their own metrics — measures that operate at the sequence level rather than the mechanism level. A chain-level metric might track end-to-end latency (how long from trigger to final output), or aggregate outcomes that only make sense when you view the full sequence.

Chain metrics are configured in the metrics panel within the chain editor and feed into dashboards like individual mechanism metrics.

What next?

• Mechanism health — understand how health states cascade through a chain.
• Dashboards — visualise chain and mechanism metrics in configurable dashboards.
• Metrics & integrations — how metric data gets into Balcony from external systems.