sfdmu-data-plans

Use semicolon-delimited field names like `Field1;Field2`, not the v4 `$$Field1$Field2` syntax. Relationship traversals such as `Parent.Field` or `GrandParent.Parent.Field` are supported and match correctly under Upsert on SFDMU 5.6.4+.

Use Upsert for objects with direct or composite unique fields and for relationship-traversal externalIds on 5.6.4+. Use Readonly for reference data loaded by another plan, Update to modify existing records, and mark empty CSVs as `excluded: true`.

This is Bug 4: `$$` composite notation fails when used in lookup reference columns, for both self-referential and cross-object lookups. Replace the composite reference with a simple single-field reference such as `ParentGroup.Code` and set the object's externalId to that simple field.

Plans are written for fresh orgs. Upsert and Update re-runs match on externalId without duplicating, but plain Insert duplicates every row and Insert with deleteOldData can be blocked by live transactional references. Load individual records surgically instead.

Only with explicit user approval and a concrete current reason, since deleteOldData deletes all existing records before inserting. On the 5.6.4+ floor the historical Upsert-matching bugs are fixed, so they no longer justify this pattern.

By default in the plan root as `<Object>.csv`. With top-level `useSeparatedCSVFiles: true`, `objectset_source/object-set-N/<Object>.csv` overrides the root file for pass N only; pass 1 always reads the plan root.