03.02 Working template

Evidence review worksheet.

Use this worksheet to prepare an agent-system review. It is a local, printable working aid: it does not submit, store, transmit, score, or decide anything. Keep sensitive information out of this page and place the completed record in your organization’s approved system of record.

Before you begin

Define the decision first: a research pilot, a production-tool connection, a data access request, a procurement shortlist, or a pause. Then use the evaluation guide to distinguish documentation from evidence. The Field Index can provide public context but does not validate an implementation.

1. Decision and scope

System, project, or proposal name

Decision to make

In scope

Explicitly out of scope

2. Evidence register

For each material claim, record the direct source, date, owner, and what the source does not establish.

Claim or question	Evidence URL or record ID	Checked on / owner	Limit or unknown

3. Authority, data, and failure boundary

What may the system read, write, call, delegate, or transmit?

Which actions require a named human approval, and how is that approval enforced?

What data classifications, connectors, retention rules, and handoffs need confirmation?

What failure, misuse, or rollback scenario still lacks an owner or tested response?

4. Evaluation and stewardship

Version and test method reviewed

What did the evaluation not test?

Accountable system owner and human reviewer

Recheck date and correction/rollback path

5. Next decision

Hold: material evidence or owner information is missing.

Narrow: reduce scope, permissions, data access, or claims before further review.

Request evidence: record exactly what source, test, owner, or limitation is needed.

Escalate: route the question to the responsible technical, security, privacy, legal, or operational reviewer.

Proceed to a separately approved decision: only after the stated evidence boundary is complete.

Decision rationale and named decision owner

Sources and limitations

The worksheet is a Coreweaver Field Index proposal. It is informed by the public context described in the evaluation guide, including NIST’s voluntary AI RMF and OWASP’s agentic threat-modeling material. It does not implement those materials, establish compliance, or replace specialist review.

Read the Field Index methodology