Learning proposals: OpenLatch proposes, an editor decides
A zone’s new Proposals tab lists four kinds of proposals, each with the evidence behind it and the recorded actions whose outcome would change: Start enforcing switches an AIP from Monitor to Enforce in a zone after 14 clean days, Add to your dictionary adds a value your reviewers agreed on, Skip repeat approvals lets an intent policy’s declarations decide, and Keep as an example keeps a reviewed action as a regression case of the intent policy. An editor accepts or dismisses each one in one click; a dismissed proposal does not come back, and History keeps every decision for 180 days. Proposals are computed from your organization’s own records, inside your organization, with no generative AI model. When an intent policy’s declarations decide, a declaration the client cannot verify is taken on the agent’s word and the action is allowed instead of the fallback; it is never recorded as verified, a matching Block rule still blocks, and an editor can revoke it from the AIP drawer. It applies on clients 0.7.2 or later; older clients keep the fallback. The AIP drawer shows each example action, editors receive a Learning notification when new proposals open, and the Public API offers the same operations in beta.