Investigation policies can select bounded saved-query work and local coverage-review tasks when admitted evidence changes. They remain subject to source rights and declared evidence limits. They cannot issue unrestricted queries, infer genealogy or write stock, quality, recall or despatch decisions.
Start from a permitted current query
An Author opens Automation policies. Under Start from a saved investigation, select the exact current-evidence run and choose Create policy draft. Access to every dataset and current Author authority are rechecked.
The draft reuses its query but has no active rules or example answers. Read the original run's starting identity, mode, direction, datasets and limits first. An as-known historic query is not the supported source for this current-policy shortcut.
Use the connected review path
Use Open in Automation when available. The shared workspace handles supported conditions, actions, independent examples and release review. If Automation workspace is not connected appears, this page does not provide an alternative native form that silently activates the draft.
Prepare explicit query profiles, start references, grouping, freshness thresholds and decisions for investigation selection and coverage review. Defaults must state no run or no task where intended. Keep source relationships and quantities unchanged.
Prove independent outcomes
Supply independently expected examples covering matching and nonmatching rules, defaults and each query profile. Expectations include the complete decision and exact native related sets and limitations, not just a successful status. Native proof runs the actual bounded investigation against permitted source cuts.
Case and whole-release reviews each require an eligible Reviewer independent of the relevant contributors, with recent authentication. The complete contributor list matters across organisational and other-product work as well as local authors. A contributor cannot activate locally to bypass review exclusion.
Interpret activation and output
Activation rechecks current source dependencies, rights, proof, review and active assignment. It does not itself produce a new investigation or any physical action. Later processing uses retained triggers and current rights. Existing saved runs keep their policy and original evidence cut.
Where processing or a trigger is stale or denied, do not claim the intended investigation happened. Inspect the saved outcome or local task that was actually recorded. Use coverage and the Automation handbook. No live model accuracy or native-source connection is established by a passing local policy proof.