Cypher self-join on the same edge type can return incorrect empty results on a populated graph #3758

@lvca

I'm running some queries on stackoverflow dataset that I shared wtih you.

We are seeing what looks like a Cypher engine bug in embedded ArcadeDB, not a client binding issue.

Environment:

• ArcadeDB 26.4.1-SNAPSHOT

A co-occurrence query over a populated property graph sometimes returns an empty result set even though the graph clearly contains matching data.

Query under test:

MATCH (q:Question)-[:TAGGED_WITH]->(t1:Tag)
MATCH (q)-[:TAGGED_WITH]->(t2:Tag)
WHERE t1.Id < t2.Id
RETURN t1.TagName AS tag1, t2.TagName AS tag2, count(*) AS cooccurs
ORDER BY cooccurs DESC, tag1 ASC, tag2 ASC
LIMIT 10

Expected behavior:

• Return the top co-occurring tag pairs.
• On our dataset this result is non-empty.

Observed behavior:

• Neo4j and LadybugDB return the same non-empty top-10 result for this query.
• ArcadeDB has produced an empty result set for the same logical dataset and query shape.

Why this looks like an ArcadeDB issue:

• Two other graph engines, Neo4j and LadybugDB, return the same result.
• The graph is definitely populated.
• A manual reconstruction against the same ArcadeDB database produces the expected non-empty top-10 result.
• In a fresh repro, the exact same Cypher query can also return the correct result in ArcadeDB, which suggests the problem may be intermittent or state/plan dependent rather than a syntax issue.

Manual validation logic:

1. Fetch all tags with their ids and names.
2. Fetch all (Question)-[:TAGGED_WITH]->(Tag) pairs.
3. Group tags by question.
4. Count unique ordered pairs (t1, t2) where t1.Id < t2.Id.
5. Sort by cooccurs DESC, tag1 ASC, tag2 ASC.
6. This produces the same non-empty top-10 result that Neo4j and LadybugDB return.

Additional note:
A different equivalent-looking formulation using collect()/UNWIND also produced obviously wrong output in ArcadeDB:

MATCH (q:Question)-[:TAGGED_WITH]->(t:Tag)
WITH q, collect(t) AS tags
UNWIND tags AS t1
UNWIND tags AS t2
WITH t1, t2
WHERE t1.Id < t2.Id
RETURN t1.TagName AS tag1, t2.TagName AS tag2, count(*) AS cooccurs
ORDER BY cooccurs DESC, tag1 ASC, tag2 ASC
LIMIT 10

In that case we observed a single row with null tag names and a large count, which also points to a Cypher execution issue around this pattern.