Span

A span is the atomic record of a single step inside an agent run — one discrete unit of work that Prefactor captures as its own record within an instance.

Spans are the raw material of governance. When Prefactor records a run, it records each step as a span: the input that went in, the output that came back, how long it took, and whether it completed or failed. Together, the spans form the activity timeline of the instance — you can read the sequence as a narrative of what the agent did and in what order. That timeline is also the basis for risk scoring: each span contributes to the instance’s peak data-risk classification by way of the risk profile assigned to the agent.

Spans can be nested — a parent span can contain child spans, with the hierarchy preserved in the activity timeline. This allows business-level actions to be recorded alongside the lower-level calls that compose them, giving reviewers both the intent and the detail in a single record.

Span types are defined by the agent’s activity schema — they can represent anything the team building the agent decides to instrument: for example user_message and assistant_message for chat turns, or separate types per tool (such as web_search and read_file) so each definition matches that tool’s inputs and outputs. Each type must have payload and result definitions in that activity schema; those can be strict or broadly permissive. If a span’s payload does not match the declared shape, the mismatch is recorded and flagged for review. The span still records regardless.

In the Admin UI

• Agent › Instance › Activity tab — the full span timeline for a single run, with inputs, outputs, duration, and status.

Related concepts

• Instance — spans belong to an instance; the instance is the unit of a run.
• Activity schema — declares the expected structure for each span type and surfaces conformance gaps.
• Risk profile — spans drive the data-risk calculation for their instance.