What the documents look like
Receiving and putaway handles the first physical-and-book confirmation when goods enter the warehouse. The working materials come from WMS, ERP procurement, supplier delivery notes and barcode scanning. Day-to-day documents include purchase arrival notes, receiving inspection sheets, putaway tasks and discrepancy reports. The slowest steps are checking arrived quantities, printing barcodes and assigning storage locations. A mismatch between physical stock and book stock leads to later picking errors and supplier disputes.
For retrieval, these materials mix identifiers, models, quantities, dates and status fields in the same text. A question can be phrased as natural language or as an exact number such as a delivery note number, a purchase order number or a material code. The two styles pull retrieval in different directions.
Constraints at the retrieval step
The knowledge base search node can link one or more knowledge bases that use the same vector model. Semantic retrieval relies on vector distance. Full-text retrieval relies on keyword hits. Hybrid retrieval merges the two paths through RRF. The minimum relevance value takes effect only under semantic retrieval or when result re-ranking is used. The reference limit is measured in tokens. Question optimization performs coreference resolution and question expansion, and it adds one model call. Image retrieval requires a corresponding image input and model capability. Re-ranking requires a re-ranking model and must be enabled in the settings.
Retrieval itself does not constitute authorization. A form collecting values does not equal business verification, approval, a database write or external synchronization. Those actions need their own downstream implementation.
Hypothetical drill sample: a warehouse receives a supplier delivery note dated 10 March 2025. The source text reads: "Delivery note number SH-2025-0310, purchase arrival note PO-2025-0228, material code M-4471, quantity due 200 pieces, quantity actually received 200 pieces, receiving clerk has signed for confirmation, pending quality inspection." A draft sentence under review reads: "This batch of M-4471 has completed receiving inspection and can be put away." The source text records only the receiving clerk's signature and the pending-inspection status. It does not record an inspection result, a putaway authorization or completed putaway.
Configuration and implementation checks
| Configuration item | Hypothetical starting value for this drill | Reason and comparison method |
|---|---|---|
| Linked knowledge bases | Link only the inventory procurement and supplier materials base that uses the same vector model | First check the vector model identifier of each base, exclude those that differ, then compare retrieval results |
| Search mode | Start with hybrid retrieval | Test the same question against semantic retrieval and full-text retrieval separately, then compare hits on identifier-type and natural-language-type questions |
| Reference limit | Hypothetical starting value of 2000 tokens | Test with arrival notes and inspection sheets of different lengths, observe content completeness, and compare against a smaller value for truncation |
| Minimum relevance | Hypothetical starting value of 0.4 | Label relevant and irrelevant samples first, then compare remaining results against 0.3 and 0.5 |
| Question optimization | Start with it off | Compare the original question with the optimized question, and check whether the subject or the document number was altered |
| Result re-ranking | Start with it not enabled | Compare ordering changes only after a re-ranking model is available and the setting is enabled |
Three common errors
- Treating a retrieval hit as authorization to execute. Input "Can M-4471 be put away?" and retrieval returns the delivery note fragment. That fragment supports only the receiving clerk's signature and the pending-inspection status. It does not support an inspection pass or an approved putaway. Check: input that question, compare whether the returned content contains an inspection conclusion field, and record the case as unsupported when only "pending quality inspection" appears.
- Using semantic similarity in place of entity and period checking. Input "arrival quantity for PO-2025-0228" and semantic retrieval may return similar descriptions from other purchase orders. Check: compare whether the returned content carries the purchase order number PO-2025-0228, exclude content that does not match, and do not treat the returned quantity as the quantity for this order.
- Linking knowledge bases that use different vector models. Input the same question and, if bases with different vector models are linked, retrieval behavior cannot be checked against the documented expectation. Check: verify the vector model identifier of each base first, exclude those that do not match, and record the actual observation. Do not run cross-model retrieval to compare missed recalls.
How to verify the setup
- For the hypothetical sample, input "What is the actually received quantity on SH-2025-0310?" Compare whether the returned content contains "quantity actually received 200 pieces" with a matching document number. The expected result is a hit on that field. If another document number or an empty array is returned, record it as a miss and do not conclude that the data does not exist.
- For the draft sentence "This batch of M-4471 has completed receiving inspection and can be put away," input "What is the inspection status of M-4471?" Compare whether the returned content contains an inspection conclusion. The source text carries only "pending quality inspection," so the expected result is unsupported. If "inspection passed" is returned without source support, record it as an evidence gap.
- For empty-result handling, check the downstream design against the paper input []. The reference content output is an array whose length can be 0, and the output path still runs when no content is found. Check whether the downstream node handles an empty array. Do not claim that a fabricated rule will return [] for a given query, and do not claim that the first hit is always returned or that an unrelated query is always empty.