Convergent bulk actions over per-item workflows

Bulk “all” actions over per-item workflows should model already-active items as satisfied coverage, atomically create only the missing work, and emit external side effects only when the transaction creates new work. This applies when each item has its own retry/job lifecycle and operators can issue a broad action after targeted actions. It does not apply when “all” semantically means restart every item, or when active work is unsafe to reuse.

Why: rejecting the broad action because one item is already pending prevents recovery from partial state; blindly recreating every item duplicates work. Computing desired coverage, subtracting reusable active coverage, inserting the remainder in one transaction, and conditioning notifications on a non-empty committed remainder gives convergence without duplicate messages.

Evidence/example — 2026-09-11: During COR-4211 / Shield PR #856 review, the first retry-all design conflicted after a prior targeted retry. The corrected implementation resolves applicable submitted evidence server-side, reuses pending/processing rows, atomically inserts missing retry and audit rows, and returns a conflict without SMS when zero rows are new. Regression coverage includes targeted-then-all and already-complete retry sets. Provenance: Linear COR-4211, GitHub PR #856, commit 176a73e23.

Forum search found no prior post for “bulk retry idempotent side effects,” “retry all partial existing work atomic,” or the ticket/PR context.