What the documents look like
Multi-warehouse allocation occurs at the order routing stage and determines which warehouse ships an order. This differs from courier handover, which happens after routing has already been decided. The work depends on WMS, OMS, ERP inventory records and platform inventory interfaces. Data updates at daily or hourly frequency. The documents handled day to day include allocation rules, inventory synchronization tables, transfer orders and stockout warnings. Product information and selection question answering needs to locate citable source text in these materials, so that answers about shipping warehouse priority, sellable inventory and transfer justification can be traced back to a specific passage.
A knowledge base search node returns cited content as an array, and the array length can be zero. The output link still runs even when no content was retrieved. When associating knowledge bases, only knowledge bases that share the same vector model can be selected together for vector search. Search parameters include search mode, citation token limit and minimum relevance. Semantic retrieval and full-text retrieval can be used separately; hybrid retrieval merges multiple result paths through RRF; result reranking uses a ReRank model for a second ordering pass. Minimum relevance takes effect only under semantic retrieval or when result reranking is used, and it acts as a filtering threshold rather than a permission or entity isolation control. The retrieval step does not perform authorization checks, and retrieval results do not indicate that business approval or a database write has been completed.
Hypothetical paper exercise sample: On 10 March 2025 a supply chain planner checks an allocation rules document. The source text reads: "Orders in the East China region are shipped first from the Suzhou warehouse; when Suzhou sellable inventory falls below the safety stock, shipping transfers to the Hangzhou warehouse." On the same date the inventory synchronization table records: "Suzhou warehouse SKU-A sellable inventory 120 units, safety stock 100 units." A proposed draft sentence reads: "Suzhou warehouse currently has sufficient sellable inventory, and all East China region orders ship from Suzhou." Compared with the source text, the source only specifies transfer when inventory falls below safety stock and does not state that all orders ship from Suzhou. The allocation rules document does not record whether this planner holds authorization to modify shipping warehouse priority; that authorization status remains unverified.
Constraints at the citation tracing step
The search node returns cited content as an array whose length may be zero, and the downstream output link continues to run in that case. Citation tracing therefore has to handle an empty array as a normal condition rather than a failure signal. Association of knowledge bases requires the same vector model across all selected bases; a base whose vector model identifier differs is excluded from joint association. The minimum relevance value is a filter threshold that applies only under semantic retrieval or when result reranking is used, and it does not separate authorized records from unauthorized ones or exclude unrelated business entities. Retrieval output is evidence of matched text, not evidence of approval, permission or a completed database write.
The hypothetical sample above sets the tracing task: compare the proposed sentence against the two source passages and determine whether the sentence is supported. The source text makes transfer conditional on falling below safety stock. The inventory table records 120 units against a safety stock of 100 units, which is above the threshold, so the condition for transfer is not met on the recorded figures. The proposed sentence states that all East China orders ship from Suzhou, which the source text does not state. The authorization status of the planner is not recorded in either document, so any statement about permission to modify priority remains unsupported.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Select the allocation rules base and the inventory synchronization table base that share one vector model | Compare vector model identifiers before joint association; exclude any base whose identifier differs |
| Search mode | Hybrid retrieval | Allocation rules contain identifiers and conditional wording; hybrid retrieval uses semantic recall and full-text recall merged through RRF |
| Citation limit | Set to 2000 tokens | This figure is a hypothetical starting point for this exercise; compare against the total length of the allocation rule passage so that one complete rule is covered |
| Minimum relevance | Set to 0.5 | This figure is a hypothetical starting point for this exercise; compare against returned rerank scores, and results below the value are filtered |
| Result reranking | Enable the ReRank model | Reranking reorders candidate text when candidates are numerous; this path does not run when no rerank model is configured |
| Cited content output | Accept the search node output array | Check each answer sentence against the cited source text; when the array length is zero the downstream link still runs |
Three common errors
- Treating retrieval output as an authorization basis. The input is the retrieval output array, the comparison object is the recorded authorization status, and the observable result is that the cited source text appears while the authorization field remains unrecorded. Retrieval returns cited content and does not determine whether the planner may modify allocation rules.
- Treating minimum relevance as a permission or entity filter. The input is the minimum relevance setting, the comparison object is whether returned results include unrelated business records, and the observable result is that the setting affects score filtering only and does not exclude unauthorized records.
- Treating an empty citation array as a process failure. The input is the search node output array, the comparison object is whether the downstream node continues to execute, and the observable result is that the output link still runs when the array length is zero.
How to verify the setup
- Input the allocation rules source text and the inventory synchronization table source text from the hypothetical sample, compare the proposed sentence "all orders ship from Suzhou" against the source wording "when below safety stock, shipping transfers to Hangzhou," and observe that the proposed sentence is not supported by the source and should be marked as pending verification.
- Input two knowledge bases with the same vector model identifier, compare whether the identifiers match, and observe that matching identifiers allow joint association while a differing identifier causes that base to be excluded and the actual observation recorded.
- Input an explicit paper-form empty array
[]as the search node output, compare whether the downstream answer step continues to execute, and observe that the downstream design should accept the empty array without treating the condition as an error termination.