The first-review objective is simple: move from “we have evidence about AI-assisted work” to “we have a Workflow Value Ledger readout a decision-maker can examine.”
What you need
Start with the minimum evidence needed to evaluate one workflow. A trace may show what happened, but it may not show what the work was for, who accepted it, what evidence was required, or whether the result became an organizational commitment.
- activity evidence: traces, chat logs, tool calls, or agent records
- context evidence: work items, acceptance criteria, policies, or business objective
- commitment evidence: approvals, delivery records, published reports, merged code, or customer-facing answers
- outcome evidence: corrections, rework, costs, incidents, or follow-up decisions
- an accountable person who can explain the workflow context
The three-step path
1. Select the workflow
Choose work that is small enough to inspect and consequential enough to govern. Define what decision, approval, customer guidance, or operational action may result.
2. Assemble approved evidence
Provide only material the organization is authorized to use. Evidence may be structured, exported, redacted, summarized, or referenced. Unknowns should remain visible.
3. Review the readout
Examine what was supported, what was missing, what remained contextual, what was withheld, and whether the workflow should proceed, be constrained, or receive further review.
What good output tells you
- what the workflow was trying to accomplish
- what evidence supports it
- what is missing or withheld
- whether commitment is ready, constrained, or blocked
- one recommended next action
When one source is not enough
Several approved sources can be reviewed together for the same workflow—for example, activity evidence plus a work item, policy, approval, artifact, or outcome record. Do not add sources just to make the bundle larger; add them only when they clarify the accountability question. Epistema keeps source identity, omissions, and withheld claims explicit.