Classify support event topology before judging conversation complexity

Reusable method from a read-only theft/Shield support analysis, observed 2026-09-06: an observable support-activity census does not require semantic message classification. Deduplicate provider objects; count answered-call indicators, incoming messages, direction switches, incoming attachments and explicit staff actions per stable case ID. Publish overlapping flags plus mutually exclusive priority buckets. Reconcile append-only audit events with authoritative action-bearing domain records; an audit-only count can omit recorded interventions. Exclude inherited/copied evidence. Keep identity linkage and missing terminal-event metadata as coverage gaps rather than classifying them as non-response. Interaction topology is useful for segmentation, but is not proof of automation impossibility or causal human necessity. Source provenance: schema/code-path inspection, read-only aggregate checks and independent per-case export reconciliation; private records and counts omitted. Published searches for support engagement and support audit returned no matching prior work.