Batch approval should preserve a decision for every item. One rejected or stale entry must not disappear inside a green batch label or cause the system to execute changes the reviewer excluded.
Batch decisions need explicit membership
A batch proposal groups related changes for review while retaining each item’s identity, revision, evidence, and consequence. The reviewer may approve selected items, reject others, or request clarification according to the supported policy.
For a proposed model, version the batch membership and each consequential proposal. Record decisions by item. Execution claims only the approved current items and returns individual completed, failed, uncertain, or skipped outcomes.
Do not interpret approval of the visible page as permission for items added later. Pagination, filters, and bulk-selection controls need clear semantics so the reviewer understands exactly which records are covered.
Test a fictional customer-update batch
An AI proposes correcting ten customer records. The manager approves seven, rejects two, and asks for clarification on one. The executor should attempt only the seven approved current proposals, preserving the other three dispositions.
Before execution, staff edit one approved record. That item may need conflict handling without preventing safe independent items from proceeding. The final report must show six completed changes and the specific unresolved exception if that is what actually happened.
| Batch condition | Required result |
|---|---|
| Item explicitly rejected | No proposed effect |
| Item added after approval | Not covered by the old decision |
| Approved item becomes stale | Item-level conflict path |
| One external write fails | Accurate partial batch status |
| Batch retry | Completed items not blindly repeated |
Use a fixture with mixed decisions and a controlled failure after several writes. Assert the exact changed-record set and absence of mutations on excluded items.
Keep partial results understandable
The batch can be operationally finished while containing denied or unresolved items, depending on the business definition. Show counts and item dispositions rather than one ambiguous success flag. Staff need to know which exceptions still require action.
Bulk review reduces clicking but can hide consequential differences. Group genuinely similar items, highlight exceptions, and provide manageable evidence. Do not use batching to bypass approval requirements that apply individually.
This batch design is proposed. The voice integration case study describes related backend work without claiming this feature. Pair it with partial integration reconciliation.
Batch approval FAQ
Does approving a batch approve future added items?
No. Bind the decision to explicit membership and proposal revisions.
Should one failed item stop every other item?
Use the business dependency policy. Independent items may proceed while dependent ones require a shared boundary.
What should a retry do?
Resume only eligible incomplete work using durable item identities and authoritative effect state.