Limitations and honest notes

• It is young. No long production history, one main author. Treat it as early-stage. If a quiet data bug would be costly, test your own queries against it thoroughly first.
• Primary keys are auto-generated strings, not sequential numbers, by default. forge fills in a UUID (or ObjectId on Mongo) when you omit id. An auto-incrementing integer key is SQL-only via f.id({ type: 'bigserial' }).
• One schema per process. createDb({ schema }) sets the active schema for the whole process. That fits one schema per service. For several different schemas at once, run them in separate processes.
• Some nested writes are partial. Deeply nested upsert, update, and set cover the common cases but not every Prisma shape.
• MSSQL upsert is not implemented in 2.3 — it throws NotImplemented pointing at v2.4 (MERGE rewrite). INSERT / UPDATE / DELETE / SELECT work today.
• Mongo cross-field geo nearTo — a near filter on field A combined with a nearTo orderBy on field B will only honor B (single $geoNear stage limit). Same field on both sides works fine.
• MultiPolygon, GeometryCollection, holes — single-polygon withinPolygon works. Multi-ring shapes need raw queries.
• 3D / Z coordinates — not modelled. Store altitude as a separate scalar.
• SRID reprojection — WGS84 only. UTM, state-plane, or other CRSes need raw queries.
• MySQL 5.7 — spatial works but without SRID enforcement; forge doctor warns.
• DuckDB — no FK enforcement (forge's app-side cascade walker handles it); no SAVEPOINT (a failing migration batch can't partially recover); no partial indexes / INCLUDE columns / ctid (replaced with rowid where needed); unique constraints cover soft-deleted rows since there are no partial indexes.
• No GUI, no plugin system. If you need a data browser or middleware, this is not that.

---