Post-Delivery Improvement

Delivery governance should improve with use. After a change ships, the run record gives the team a place to inspect what happened, compare outcomes with the rule, and decide what to change next.

Learn from the run, not a retrospective guess

Farna Control keeps the pipeline snapshot, node outputs, evaluation results, gate decisions, cost, and final artifact with the run. A team can inspect where a workflow paused, which check rejected an output, how much a run cost, and whether the connected system reported the expected result.

Run-level evidence is available in the Community tier. Organization-scale analytics, retention, and export depend on the plan and configuration of the deployment.

Turn observations into a safer next version

Evaluation definitions and pipeline versions make changes explicit. A team can adjust a prompt, schema, model backend, connector binding, or gate, then compare later runs with the earlier record. Autonomy can stay at manual approval while the team gathers evidence. Promotion is a decision by the authorized operator or owner team, not an automatic reward for a passing run.

Cost reports and per-run accounting add another improvement signal. If a workflow becomes slower or more expensive, the run and node records show where to investigate. The model provider and connector systems remain authoritative for the external events they report.

Current and planned surfaces

Run inspection, snapshots, evaluation gates, cost records, and pipeline versioning are current platform capabilities. Organisation-scale analytics and retention depend on the plan. A future workflow can connect more external feedback systems, but Farna Control does not claim to provide a universal post-delivery feedback loop today.

Where to go next

• AI Agent Cost Visibility: spend and run-level cost signals
• Pipelines: versions, snapshots, and comparison
• Evals: checks that inform autonomy decisions