Readiness needs more than a score
A high average can hide a missing permission boundary or an untested rollback. Assign critical gates to the dependencies that cannot responsibly remain unknown for the intended pilot. Keep optional improvements separate from those blocking gaps.
What to ask for
A task owner, authorised data access, action limits, representative evaluation, security review, an incident route and a recoverable release are useful starting points. Their precise meaning depends on the application and the organisation. NIST’s AI Risk Management Framework provides an organising reference; checking a box on this site does not establish conformity with it.
From checklist to release record
The Evidence Planner records ready, partial or unknown status with local notes. Its download is a discussion record, not a certificate. Attach actual test results and approvals in your organisation’s controlled process before release. Revisit the evidence after a changed model, connector, permission or operating boundary.