v3.8.1

Fixed

• detect_site_outage now correctly enumerates the collectors serving a device group. Previously, _collector_ids_from_devices looked for currentCollectorId / preferredCollectorId on device records, but the formatted get_devices sub-tool response renames that field to collector_id. The mismatch meant the CollectorDown signal was always zero for the composite when called through the normal dispatch path. Surfaced during the v3.8.0 smoke test on Ryan's portal (500 devices in group 1 reported collectors_serving_group: 0). The helper now checks collector_id first, with fallback to the raw API field names so raw-data callers still work.
• _count_dead_devices now counts LM hostStatus=2 (dead-collector) in addition to hostStatus=1 (dead). Both indicate devices not reporting metrics; counting only one under-reported device silence. Also reads the formatted-response status field before falling back to the raw hostStatus, matching the shape that arrives from the composite's sub-tool dispatch.

Added

• Six direct unit tests for the _collector_ids_from_devices and _count_dead_devices helpers covering formatted shape, raw API shape, absent fields, and alertStatus-only fallback.

Verified

• Full 1779-test suite passing (6 new helper tests).
• Ruff check + format clean.