A single instance can be anecdotal. These patterns repeat — the same sequence of entity types and the same sequence of relationship types, seen N times across N distinct cluster contexts. The scanner walks 2-hop and 3-hop paths nightly and groups identical sequences. Severity scales with count.
The scanner walks all 2-hop and 3-hop paths in the connections graph nightly. Each path is classified by the sequence of entity types (e.g. company → person → elected official) and relationship types (e.g. lobbies → committee member).
Identical sequences are grouped. Groups with 3+ distinct instances are surfaced as patterns. Severity scales with count: Critical (≥20), High (≥10), Elevated (≥5), Notable (≥3).
Instance deduplication ensures A→B and B→A are counted once. Each pattern stores up to 50 representative instances for inspection.
