Stamp scope onto every live item in item_ids and materialise the derived splits. Per-item semantics are exactly :func:set_item_distribution_link's — 'all'/'engineering' link (an unavailable scope is accepted; the sync no-ops until the aggregate exists), 'none' is the explicit opt-out that keeps the last computed split and is not recaptured by a future subgroup link, and NULL (follow-subgroup) is never written. An item already following via its subgroup is still stamped: the selection is an explicit per-item instruction, so the link outlives the subgroup's.

Mirrors :func:bulk_set_item_apportionments_by_ids for auth/scoping: the claim_id-scoped query is the authorization boundary (ids on another claim match nothing). Synthetic salary ids and real salary-category rows are skipped (employees are apportioned on the Salary tab), as are already-removed, asset-promoted, linked refund (derived — they follow their original via the mirror), and unknown ids. updated_count counts items whose stored scope actually changed, so re-applying the same scope reports 0 — except under 'none', which also counts rows whose R&D it zeroed (an already-opted-out row can still be carrying a derived split to clear).