What the documents look like
Closure settlement runs in the opposite direction from site selection and store opening. It handles exit and asset recovery rather than expansion. The working data comes from POS inventory, the fixed-asset system, lease contracts and the finance ledger, and it updates irregularly, once per store. The everyday materials are closure approvals, inventory clearance forms, lease termination agreements and asset count sheets. These materials sit in different states at the same time. An approval may be pending review. A termination agreement may be signed with no effective date recorded. A count sheet may cover only part of the assets.
Workflow routing has to read those states as they actually are. A pending review is not an approval. A signature is not an effective contract. Operating reports and anomaly reviews carry each state into node inputs unchanged so a human can compare them. The materials listed here are the ones this page works with; other closure documents are not described and remain unknown.
Constraints at the workflow routing step
The Form Input node and the User Selection node are both user-interaction nodes. When either is triggered, the conversation enters an interaction state, the workflow state is recorded, and execution continues after the user finishes interacting. Form Input collects required information and passes it to later nodes. User Selection judges the user's choice and runs the corresponding branch. The recorded interaction state is the node's own behaviour; the documented material describes no separate global state switch.
Information collected about a store, a material and its state must be passed explicitly to later retrieval, judgment or interface nodes, and output must be compared item by item against input. Order writing, document backfill and external status synchronisation require a separate business interface or a later process. A form or a retrieval result does not constitute authorised execution. Knowledge-base retrieval keeps the conditions stated in the documentation, and a relevance score carries no permission or entity-isolation duty. Libraries built on different vector models cannot be associated together; check the identifier first and exclude those that do not match.
Configuration and implementation checks
The table below works from a hypothetical paper exercise sample. The sample source text is: "Closure approval A-001, store S-01, status pending review; lease termination agreement B-002, signed 2026-03-01, no effective date recorded; asset count sheet C-003, scope covers first-floor equipment." The draft sentence to be judged is: "S-01 has completed lease exit and the deposit can be settled."
| Configuration object | Setting or check | Link to the sample |
|---|---|---|
| Form Input node | Make store number, material number and material status required fields | Collects S-01, A-001, B-002, C-003 and each status |
| User Selection node | Offer "continue review" and "supply more materials", each connected to its branch | Routes the "completed lease exit" judgment to the supply-materials branch |
| Interaction state check | Triggering enters the interaction state and records workflow state, waiting for the user | Checks whether the judgment is waiting in the supply-materials state |
| Output handover check | Pass form information explicitly to the later judgment node and compare input with output | Compares whether "pending review" and "no effective date recorded" pass through unchanged |
| External action split | Deposit settlement and contract effectiveness confirmation attach to a separate business interface or later process | The settlement in the draft sentence is not executed by form submission |
Three common errors
- Treating "pending review" as approval. In the sample, closure approval A-001 has status pending review. If a node judges approval passed on that basis, the input is the A-001 status field, the comparison is "pending review" against "approved", and the observable result is that the two values differ, so the judgment does not hold.
- Treating "signed" as an effective contract. In the sample, lease termination agreement B-002 has only a signing date and no effective date recorded. If the draft sentence says the lease exit is complete, the input is the B-002 date field, the comparison is "signed" against "effective", and the observable result is that effectiveness evidence is missing, so the judgment remains unverified.
- Treating retrieval or form submission as authorised execution. In the sample, the draft sentence says the deposit can be settled. The input is the S-01 information collected by the form, the comparison is information collection against settlement execution, and the observable result is that no settlement interface is connected to the node, so the action is not executed.
How to verify the setup
- Check the draft sentence "S-01 has completed lease exit and the deposit can be settled" clause by clause. The input is A-001 pending review, B-002 with no effective date recorded, and C-003 covering first-floor equipment only. The comparison is the completed lease exit stated in the draft against the actual material states. The observable result is that approval has not been granted, the effective date is missing and the count scope is partial, so the sentence is unsupported and the supply-materials branch applies. This is the expected outcome of the paper exercise, not a recorded test.
- Use the paper input
[]to check empty-result handover. The input is an empty retrieval result, and the comparison is whether the downstream node is designed with a branch for an empty result. The observable result is that the downstream node does not treat the empty result as authorised or settled, so the handover is correct. No claim is made that a fictional rule retrieval necessarily returns empty. - Check cross-model identifiers. The input is the vector model identifier of two knowledge bases, and the comparison is whether the identifiers match. The observable result is that different identifiers exclude joint association, and the actual observation is recorded without running a cross-model retrieval to compare missed recall.