Reviewing Agent Runs
Every time an agent works on a record, Full Fabric keeps a run: when it happened, who it was about, what the agent concluded and how it got there. Runs are how staff check an agent's work, find out why a result is missing, and see what was sent to the AI model.
Opening the Runs
In Settings → Automation & AI → Agents, open the menu on an agent's row and select Runs. The page lists the agent's most recent 200 runs, newest first. Search by name or email narrows the list to runs about matching people.
| Column | Shows |
|---|---|
| Timestamp | When the run was created, with the time below the date |
| Profile | The person the run is about, with their email below |
| Target | The record the agent worked on: the application, the person, or the uploaded document |
| Status | Where the run stands (see below) |
| Trigger | The event that started the run |
| Result | The score, or the document verdict |
| Justification | The agent's explanation of its result |
| Error | Why the run failed, when it did |
Run Statuses
| Status | Meaning |
|---|---|
pending |
Queued, waiting for a worker. Further events for the same record join this run |
running |
The agent is working on the record |
succeeded |
The agent finished and its result is saved |
failed |
The agent stopped; the Error column says why |
A run that fails is tried again automatically, up to three attempts in all. Each attempt appears as its own row.
Common error messages:
| Error | What to do |
|---|---|
| Blocked by the institution AI governance | AI model access is off. Turn it on in Governance and enable the agent again |
| AI provider connection problem. Check provider credentials with the administrator. | The active provider's credentials were rejected. Check the provider in Governance |
| (error type): agent run failed | An unexpected failure. Contact support with the run's timestamp |
Some events start no run at all: when the agent was disabled before its turn came, when the record no longer exists, when the agent's templates or programmes no longer cover the record, or, for AI-capable agents, when AI model access is off.
Run Details
Select a row to expand it.
- AI-capable agents show the System prompt sent to the model (your prompt plus the response format Full Fabric adds), then each exchange with the model as an Iteration with its Request and Response.
- The document validator shows each step it took as an Iteration, with its outcome and duration in milliseconds, and the Request and Response of that step. It records how the document was read and the verdict, never the text read from the document.
- Potential scorer runs show the Suggested potential for each journey first (see below).
"No audit entries recorded" means the run has nothing to show yet, usually because it is still pending.
Where Results Appear
Each agent keeps one result per record: a new run replaces the previous result for the same record. Results are deleted with the application they belong to.
| Agent | Where staff see the result |
|---|---|
| Application scorer | A Score and a Justification column per agent, named after the agent, in application lists and exports; filters on the score in segments; the score on each application |
| Potential scorer | Its score and justification next to the person's applications in application lists; the potential on the journey card |
| Document validator | The verdict badge on the upload, the proven details and the validation status filters. See Passport/ID Field |
Scores are shown with two decimals.
Note: Agent results are visible to signed-in staff wherever the underlying record is visible, for example as columns in application lists and badges on uploads. The output of an AI-capable agent is stored for staff review: it never moves an application or a person to another state, and never messages an applicant. The one exception is the potential scorer, which sets journey potentials, an internal field staff can change back.
Potential Suggestions
The potential scorer applies its suggestions itself. A suggestion it could not apply stays open for staff.
In the run details, Suggested potential lists each journey with the change (for example Medium → High) and the agent's one-sentence reason. An open suggestion has an Apply button; an applied one says Applied by the agent or Applied by name, with the date and time.
On the person's profile, the journey card shows open suggestions as "AI agent's potential suggestion: High", with an Apply button for staff who can change the journey. Once a potential comes from the agent, the card says "Set by AI agent." or "Applied by name from AI agent's suggestion.", with the date. The AI agent link opens the agent's runs.
| Rule | Behaviour |
|---|---|
| Latest run wins | Only the suggestions from each agent's most recent successful run on the person are offered |
| Stale suggestions | A suggestion is no longer offered once the journey's potential has changed. Applying one that went stale returns "The journey's potential changed since this was suggested." |
| One apply | Two staff members cannot apply the same suggestion; the second gets "Already applied." |
| Reverting | Staff change the potential back from the journey card like any other potential |
Every potential the agent sets is recorded on the person's change history, signed as "AI agent: agent name".