What the documents look like
On-demand retail store fulfillment processes online orders by the minute, which differs from periodic stocktaking. Procurement and supplier records therefore carry three traits. First, documents are generated per order: purchase requests, supplier delivery notes, out-of-stock reports and delivery exception slips move between the store and the backend. Second, the supply status of one supplier can differ across stores, so records must carry three subjects: store, supplier and item. Third, document status changes quickly, and pending review, reviewed and effective are distinct facts. These traits directly affect knowledge base retrieval. If the query lacks a store or supplier subject, recalled content may come from another store's records for an item with the same name. If a record treats a quotation validity period as a supply term, the retrieved citation can mislead a later procurement judgment. The knowledge base search node can associate one or more knowledge bases that use the same vector model; bases with different vector models cannot be associated together.
Constraints at the retrieval step
Search runs against the whole knowledge base; a collection is only for classification management and does not affect search results. Semantic retrieval relies on vector distance, full-text retrieval relies on keyword hits, and hybrid retrieval merges the two paths through RRF before ReRank performs a second ordering. The minimum relevance score takes effect only under semantic retrieval or when result reranking is used; it is a filtering threshold and does not perform permission or entity isolation, and retrieval itself does not constitute authorization. The citation limit is measured in tokens, not a fixed number of items. Question optimization performs coreference resolution and question expansion and adds one model call.
Hypothetical drill sample: the source procurement record of a supermarket home-delivery store reads, "Supplier A, delivery scope includes this store, quotation validity period for Item A ends 30 June 2025, current supply status normal." A draft sentence to be judged reads, "Supplier A's supply term for Item A at this store ends 30 June 2025." The judgment points are the same subject, the same period and the same field. The source states only a quotation validity period and does not state a supply term, so the draft sentence rewrites the quotation validity period as a supply term and is unsupported. If the draft sentence read, "Supplier A's quotation validity period for Item A at this store ends 30 June 2025," it would match the source. A purchase request pending review does not show that approval was granted; signing, reaching cooperation and contract effectiveness are three distinct facts.
Configuration and implementation checks
Configuration is tied to this page's sample and lists only documented settings or implementation checks.
| Configuration item | Value or check on this page | Basis and explanation |
|---|---|---|
| Associated knowledge bases | Select procurement and supplier record bases that use the same vector model | The node can associate one or more knowledge bases with the same vector model |
| Search mode | Compare semantic, full-text and hybrid retrieval against the test question | No single mode is presumed suitable for all records |
| Citation limit | Set the hypothetical trial starting value at 2000 tokens and compare with 1000 and 4000 for content completeness | This value is a hypothetical starting point, not an official default; the unit is tokens |
| Minimum relevance score | Label relevant and irrelevant samples first, then adjust the threshold | Takes effect only under semantic retrieval or when result reranking is used |
| Question optimization | Compare the original question with the optimized question and check whether subjects were altered | Used for coreference resolution and question expansion |
| Result reranking | When a rerank model exists and is enabled per settings, check candidate ordering | The reranked score is 0-1 and can be used for filtering |
Three common errors
- Treating a quotation validity period as a supply term. The input is the source text "quotation validity period ends 30 June 2025" and the draft sentence "supply term ends 30 June 2025." Comparing field names shows the draft contains a field the source does not state, so it is judged unsupported.
- Treating a purchase request pending review as approved. The input is the status field "pending review" and the draft sentence "this purchase request has been approved." Comparing status values shows they do not match, so it is judged unsupported.
- Treating a retrieval hit as authorization. The input is a supplier record citation returned by retrieval and the draft sentence "this supplier has obtained procurement authorization for this store." Comparing the authorization field shows the record does not state authorization, so it is judged to require verification.
How to verify the setup
- For the sample source and draft sentence, input the source text, the draft sentence and the judgment requirement, then compare field names and field values. The condition is met when the draft field matches the source and the value matches; a field the source does not state is unsupported.
- For the pending-review status, input the status field "pending review" and the draft sentence "approved," then compare status values. The condition is met when the two match; when the values differ, the judgment is unsupported.
- For empty-result continuation, input the paper retrieval result
[]and compare whether the downstream node continues with an empty array. The condition is met when the citation content length is 0 and the path continues. For bases with different vector models, check identifiers first and exclude non-matching ones; do not run a cross-model retrieval to compare missed recalls.