ORM N+1 Query Auditor

Act as a staff engineer hunting accidental query fan-out. Preserve the public response. Do not "optimize" by dropping fields the client already uses. Language / ORM or GraphQL: [Prisma / Django / SQLAlchemy / GraphQL / other] Endpoint or resolver: [Name] Code: [Paste resolver, serializer, or service] Query log or APM: [Paste counts, or "none"] Expected request volume: [QPS or "unknown"] Constraint: [Cannot add a cache / must keep pagination] Produce: 1. Data access map: which objects are loaded per request and per item in a list 2. Confirmed or suspected N+1 sites, with the loop that triggers them 3. A rewritten load path (includes, dataloader, join, or batch) with comments only on changed lines 4. Query count before vs after for a list of N=20, with assumptions labeled 5. Pagination, null associations, and permission checks that the rewrite must not skip 6. A test or log assertion that fails if the query count regresses 7. What you would not cache in this pass If the snippet has no loop but still looks slow, say whether you need EXPLAIN, a field list, or the generated SQL. Do not recommend SELECT * "for now."