What the documents look like
Closure settlement moves in the opposite direction from site selection and store opening. The work is exit and asset recovery, so the operating review reads records that describe wind-down rather than launch. The main data comes from POS inventory, the fixed-asset system, lease contracts and the finance ledger. Updates happen irregularly and store by store, which means a review conclusion may rest on a document that was current at one point and has since been superseded. The document surface for this page is the closure approval, the inventory clearance form, the lease termination agreement and the asset count sheet. The most time-consuming steps are counting inventory and assets, negotiating the lease exit, and settling employee and franchise amounts. Citation tracing matters because a review statement has to return to a specific material, show that material's state, and stay inside the access boundary the reviewer actually holds. Without that, deposit loss, asset loss and labour or franchise disputes become harder to reconstruct.
Constraints at the citation tracing step
The knowledge base search node can select one or more knowledge bases that share the same vector model. Bases built on different vector models cannot be associated together. At the check step, confirm the identifier first and exclude bases that do not match; do not run a cross-model search in order to compare missed recalls. Citation content is output as an array and its length can be zero, so an empty array is not the same as a workflow error that terminates the run. The search output still has to be handed to the answering step, and each statement must be checked against the cited source text.
The minimum relevance value is a filter threshold. Semantic recall and the relevance score do not carry permission or entity isolation duties, and retrieval itself is not authorization. Hybrid retrieval merges multiple result paths through RRF, and reranking is an optional path that requires a rerank model and must be enabled according to the setting. Image retrieval requires an image input and the corresponding model capability. These paths do not mean every request executes all of them.
Hypothetical drill sample. Source material one reads: "Store A closure approval form states: applicant operations specialist, submission date 10 March, status pending review." Source material two reads: "Lease termination agreement states: lessor and lessee have reached an intention to terminate for Store A, dated 12 March, no signature block for either party is present." The draft statement to be judged reads: "Store A has completed closure approval and has terminated the lease." The authorization boundary for this request is: checking may only occur inside the closure settlement material search scope to which the reviewer has access; anything outside that scope is recorded as an evidence gap.
Configuration and implementation checks
| Implementation check | Setting or action on this page | Relation to the sample |
|---|---|---|
| Associated knowledge bases | Select only knowledge bases sharing the same vector model | The sample materials must fall inside the selected bases to be checkable |
| Citation output | Accept the array, allow length zero | An empty array is treated as nothing retrieved, not as an error |
| Citation handoff | Compare each statement against the cited source text | The draft statement must return to the two source passages |
| Result reranking | Enable according to the setting and read the 0-1 score | The score is used for filtering, not as a basis for authorization |
| Citation ceiling | For this hypothetical trial the starting value is set at 3000 tokens, chosen because the two passages are short and allow sentence-by-sentence checking; compare against the configured citation ceiling value to observe whether truncation occurs | Truncation would drop the status field |
| Material currency | Test with the current version and a historical version of the documents | Different versions can support different conclusions |
Three common errors
- Treating "pending review" as already approved. The input is the status field on the approval form, the comparison is between the two mutually exclusive values "pending review" and "approved", and the observable result is the actual value of the status field. In the sample that field reads pending review, so the approval part of the draft statement is not supported.
- Treating "intention to terminate reached, no signature block present" as a terminated lease. The input is the agreement date and the signature block, the comparison is between an expression of intention and a fact of signature, and the observable result is whether the signature block records signatures. In the sample no signature appears, so the termination part of the draft statement is not supported.
- Using a similarity score to infer permission or entity isolation. The input is the minimum relevance setting and the returned entries, the comparison is between the score and the subject each entry belongs to, and the observable result is whether the entry falls inside the authorized scope. A high or low score does not indicate that a record has been authorized.
How to verify the setup
- Input the sample approval passage and the draft statement, compare the approval form status field against "approved", and observe the value of the status field. The approval part of the draft statement is supported only when the field explicitly reads approved.
- Input the sample agreement passage, compare whether the signature block records signatures by both parties, and observe the content of the signature block. The termination part is supported only when signatures are recorded; in the sample that part remains to be verified.
- Input an explicit paper-level empty array
[]together with the downstream node design, compare whether the downstream continues executing on an empty citation, and observe whether the downstream raises an error. This check verifies the empty-result handoff design and does not claim that a fictional rule search will necessarily return an empty array.