0.8.0-dev7: :AFFECTS edge resolution + target/subject consistency

[stevenca]

Summary

Second small polish to the dev5 pipeline, caught by deployment verification.

The dev5+dev6 deploys produced 31 :ReflexEvent nodes (real data, including 28 from Meraki MS switches after the dev6 fix landed), but none of them connected to their :Device nodes via :AFFECTS.

Root cause: the link_down handler preferred payload[\"device_id\"] over payload[\"device\"], which produced targets like 'meraki:Q4CD-Y6FW-EKVS|Port 9' — a Neo4j node-id form that no Device node had as its name or platform_id.

Fix (two-sided, low-risk)

1. Handler side: link_down now prefers payload[\"device\"] (human name) over payload[\"device_id\"]. Falls back to device_id when device is absent (legacy publishers, traps).
2. Sink side: Neo4jReflexEventSink AFFECTS-edge MATCH now resolves on three identifiers: d.name, d.platform_id, AND d.id — the third catches node-id-form targets if a future publisher emits them.

The handler fix makes the target consistent with the subject token; the sink fix makes the edge resolution forgiving of any remaining identifier-form drift.

Tests

• 2 new cases in `tests/reflex/test_handlers.py` covering both ordering preferences.

Backfill plan

After deploy, run this Cypher once to retrofit the 31 already-captured events:
```cypher
MATCH (e:ReflexEvent) WHERE NOT (e)-[:AFFECTS]->(:Device)
WITH e, split(e.target, '|')[0] AS device_key
MATCH (d:Device)
WHERE d.name = device_key OR d.platform_id = device_key OR d.id = device_key
MERGE (e)-[:AFFECTS]->(d)
RETURN count(*) AS backfilled
```

Test plan

• CI: all 8 jobs pass
• Deploy
• Run the backfill Cypher
• Confirm `MATCH (e:ReflexEvent)-[:AFFECTS]->(d:Device) RETURN count(*)` returns 31+
• Verify next genuine SNMP-detected flap lands with its `:AFFECTS` edge already attached

Made with Cursor