What Farna Control governs

Farna Control governs consequential automated changes and decisions, together with the evidence needed to accept, reject, correct or expand them.

The governed object is not an abstract model or a private chain of thought. It is the delivery action: its input, the versioned pipeline and prompts that shaped it, the agents and tools it used, the policies and gates applied, the decision made, and the resulting artifact.

The boundary matters

Farna Control governs

Consequential software-delivery changes, approvals, policy decisions, handoffs and the evidence that lets a team inspect and challenge them.

Farna Control does not govern

An agent's private reasoning.
Model internals or a provider's hidden process.
Every external action in the systems you connect.

This boundary is deliberate. Farna Control records and controls the decisions that matter to delivery without pretending that it can observe or certify everything that happens inside a model or outside the instance.

Evidence keeps its provenance

A trustworthy record distinguishes what the system saw from what it calculated, what someone claimed and what a party attested.

Observed

Recorded directly by Farna Control: a run started, a gate was decided, a connector returned a response or an artifact was produced.

Derived

Calculated from observed records by a documented rule: a duration, a score, a chain verification result or a policy outcome.

Reported

Claimed by an agent, connector or operator. It remains a claim, so a reviewer can see who reported it and what evidence supports it.

Attested

Signed or otherwise vouched for by an identified party outside the ordinary run record. The attestation is retained with its scope and provenance.

Autonomy is a progression, not a switch

A team earns a larger operating boundary by making the controls explicit and looking at the evidence from the smaller one. These are the progression stages; the product surface for a stage is labelled according to what has shipped, not what is planned.

1
Manual
A person starts and decides the work.
2
Notify-on-complete
The system completes the configured work and tells a person what happened.
3
Human review
A person must review a consequential result before the pipeline can continue.
4
Bounded autonomy
The pipeline can act within explicit limits, budgets and policies, with exceptions routed to a person.
5
Earned autonomy
A team expands the boundary when evidence from evaluations, history and review supports doing so.

Evidence by edition

Community

Run-level evidence is free: inspect a single run, its inputs and outputs, the decisions taken and the artifact produced. The governance boundary is not paywalled at the point where a person needs to understand one consequential change.

Team

Team adds organization-scale controls and evidence surfaces, including the features needed to compare, retain, export and administer governance across teams and runs.

The practical test

For every important automated change, ask: what was observed, what was derived, what was only reported, who attested to it, which control applied, and what evidence would let us accept, reject, correct or expand the action?