What the documents look like
Closure settlement handles exit and asset recovery, the reverse direction from site selection and store opening. The working materials come from POS inventory, the fixed-asset system, lease contracts and financial ledgers. Day-to-day documents include closure approvals, inventory clearance forms, lease termination agreements and asset count sheets. Chain operations, finance, legal and the landlord contact each hold parts of this material. Updates are irregular and each store is handled separately.
Contract and transaction material comparison requires one store, one period and one field to share a single meaning. The retrieval step only locates candidate source text. It does not confirm whether an approval passed or whether a payment was settled. A closure approval sitting in a knowledge base is a document, not an authorization state; a termination agreement that has not been signed is not a completed termination; an offered settlement option is not an executed payment.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases built on different vector models cannot be associated together. Check the model identifier on each base first and exclude those that do not match. Do not run cross-model retrieval in order to compare missed recalls.
Search modes are semantic search, full-text search and hybrid search. Semantic search relies on vector distance and suits natural-language description. Full-text search relies on keyword hits and suits numbers, model codes and proper nouns. Hybrid search uses both recall paths and merges them through RRF. The minimum relevance value takes effect only under semantic search or when result re-ranking is used; it is a filter threshold and does not perform permission or entity isolation. The reference limit controls how many tokens of content each search may cite at most. Question optimization performs coreference resolution and question expansion before retrieval, which helps in multi-turn dialogue and short questions. Image retrieval requires an image input and the matching model capability. Re-ranking requires a model and runs when enabled by configuration. These optional paths do not all execute on every request. The reference content is output as an array whose length can be 0, and the downstream chain still runs when nothing is found.
Hypothetical paper exercise sample: among the settlement materials of one store of a chain brand, the lease termination agreement states "Party A agrees that Party B will complete the handover by 31 March 2025, and the deposit will be refunded within fifteen working days after handover acceptance." The asset count sheet states "The store's freezer, number A-07, is recorded in the fixed-asset system at book original value, and its count status is pending disposal." The inventory clearance form states "As of 20 March 2025, the store's remaining inventory has entered the clearance process, and outbound confirmation is not yet complete." A draft sentence reads "The store's deposit has been refunded and freezer A-07 has completed disposal." The task is to judge whether the draft sentence has material support.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this exercise | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the same-vector-model bases holding closure approvals, inventory clearance forms, lease termination agreements and asset count sheets | Check the vector model identifier of each base first, exclude those whose identifiers differ, and do not run cross-model retrieval |
| Search mode | Run the sample question with hybrid search first | Compare item by item against semantic search alone and full-text search alone, and observe the hit differences between number queries and natural-language queries |
| Reference limit | Hypothetical starting point set to the token volume needed to cover one lease termination agreement and one asset count sheet | Compare with actual materials of different lengths to see whether returned content is complete, then adjust; this is not an official default value |
| Minimum relevance | Hypothetical starting point at a middle value, used only under semantic search or result re-ranking | Label relevant and irrelevant samples first, then compare whether irrelevant fragments decrease as the threshold changes; this is only a starting point for this exercise |
| Question optimization | Enabled for short sentences such as "the store's deposit has been refunded" in the sample | Compare the original question with the optimized question and check whether the subject was altered; it can be turned off when the question is already clear |
| Result re-ranking | Observed when a re-ranking model is available and enabled by configuration | Compare candidate ordering before and after re-ranking, confirm the re-ranking score can be used for filtering, and make no inference when it is not enabled |
Three common errors
- Treating a retrieval result as an authorization conclusion. In the sample, the lease termination agreement states only the handover deadline and the condition for deposit refund, while the draft sentence writes "the deposit has been refunded", turning a pending condition into a completed fact. Check whether a refund voucher or an acceptance record exists in the material, and record the item as unsupported when it is absent.
- Merging different fields into one judgment. The asset count sheet states that freezer A-07 is "pending disposal", while the draft sentence writes "disposal completed". Disposal status and count status are different fields. Check whether a disposal completion record exists; a count status alone cannot establish that disposal is complete.
- Treating stage evidence as a negative conclusion. The inventory clearance form states "outbound confirmation is not yet complete", which is stage evidence. It cannot be used to conclude that the inventory does not exist or that clearance failed. Record the current value of that field and keep the evidence gap open.
How to verify the setup
- Input the lease termination agreement text from the sample and the draft sentence "the store's deposit has been refunded". Compare whether the material states a completed refund field or a completed acceptance field. When only the handover deadline and the refund condition appear, judge the sentence unsupported; do not judge it as refunded.
- Input the asset count sheet text and the draft sentence "freezer A-07 has completed disposal". Compare whether the count status field and the disposal completion field are the same field and whether their values are mutually exclusive. When the count status is "pending disposal" and no disposal completion field is stated, judge the item as pending verification.
- Input the inventory clearance form text and compare "outbound confirmation is not yet complete" against any outbound completion record. When no outbound completion is stated, record it as unsupported; do not use that entry to judge that inventory clearance failed. For the empty-result path, use an explicit paper input of
[]to check the downstream design, and do not claim that a fictional rule search necessarily returns[].
This page addresses only chain-store closure settlement, contract and transaction material comparison, and knowledge base retrieval and recall. Which materials, operations and requests sit in which authorization state during an operations handover cannot be treated as authorized execution by retrieval or by a form.