Record Guide
How to read run records and metrics is the same as in the SaaS edition (Record Guide, Records and metrics). This page sets out what differs in the Marketplace edition.
Reading a single run
A run record shows “Runtime image” in addition to the SaaS edition's items (under “Run info” ＞ “Settings at run time”). It is the container image of the registered agent that ran in that execution. In a workflow it shows the image digest an administrator certified (sha256:…); in a cycle, the registry, image name and tag.
From a workflow's run record or Workflow history, “Case details” opens the case screen. The versions of settings and artifacts are seen on the case screen (Case Guide).
What “View current settings” in the heading shows differs from the SaaS edition. For a registered agent whose contract is registered and can be read, it shows a summary of the contract (“Effect on external systems” with the execution limit, “Files”, “Outputs” and “Connections”). For an agent with no contract, or one whose contract cannot be read (such as an agent used in chat that you place only in a cycle), it shows “This information is not available yet”. In neither case is “Edit Agent” shown.
When it could not be launched
When a registered agent could not be launched, the run record shows the reason. Here are the reasons that often appear, and what to fix.
| Reason | What to fix |
|---|---|
| The Agent is not certified for this project's execution rules | Compare the project's basic rules with the agent's “Supported run rules” (Execution Rules and Traffic Guide) |
| The Agent's certification details were incomplete | An administrator registers the agent's certification again |
| The Agent contract was invalid | An administrator fixes the agent's contract |
| The linked service selected for the step was unavailable | An owner checks the connections in the step settings and the connectors in Settings |
| The approved version of the linked service changed or was not found | The connector's settings changed. An owner chooses the connection again in the step settings |
| The credentials for the linked service were missing or no longer valid | Reconnect the connector |
| The Agent image to run could not be identified | An administrator checks the agent's image settings (registry, image name) (mainly cycles) |
| Environment variables set by the administrator conflicted | An administrator removes names the platform settles from the agent's environment variables (cycles) |
| The step's environment variables included a name that cannot be used | An owner removes the name that cannot be used from that step's environment variables (cycles) |
The cycle result “Run ended (completion not measured)”
Even when a cycle's registered agent finishes its processing and exits normally, the platform cannot judge whether it did its job. The result of such an execution is “Run ended (completion not measured)”.
- It is counted as neither a success nor a failure. It is not part of the denominator of metrics such as the agent completion rate. The Overview metrics show the number with “Completion not measured” separately
- The result filter in Execution history has “Run ended (completion not measured)”
- What the execution did is checked in that execution's conversation
If the agent is built to report whether it completed (the SDK's report_completion), it is recorded with the same seven results as in the SaaS edition. How to build it is in Your First Agent.
When the agent's process ends abnormally, the Pod disappears, or the execution limit is exceeded, any report from the agent is not used, and it is recorded as failed or timed out.
Stopping a cycle execution
A cycle execution in progress can be stopped with “Stop” in its run record. It is shown to owners only. It can be used on executions waiting to start and on executions that are running.
Stopping from the conversation screen does not stop a cycle execution. Use “Stop” in the run record.
How the metrics are counted
The metrics are counted as in the SaaS edition (Run Results Reference). The denominator is, in a workflow, the executions of registered agents placed on steps (excluding review steps and steps passed through), and in a cycle, the sweep executions. “Run ended (completion not measured)” is not part of the denominator.
Records (audit trail)
In addition to the SaaS edition's events, Records (audit trail) keeps “Case Config Changed”. It is the record made when the case settings are saved as a new version. Events about consulting on improvements, email and lanes do not exist in the Marketplace edition.
Revisiting it from the track record
Of “Revisiting it from the track record” in the SaaS edition's Record Guide, the feature for consulting the AI on improvements and carrying them out does not exist. To revisit, read the run records and the track record of reviews, and choose where to fix as follows.
| What you see | What to fix |
|---|---|
| The values in the step settings do not fit | An owner fixes the step settings |
| The same correction is made at every review | Consult an administrator about the contract's defaults, and the developer about how the agent is built |
| Failures and timeouts continue | Have the developer check how the agent is built and its execution limit, and whoever operates the environment check the traffic allowed (Execution Rules and Traffic Guide) |
| “Not allowed by rules” continues | Compare the project's basic rules with the agent's “Supported run rules” |