What the documents look like
Marketplace promotion enrollment and price management turns on price, inventory and campaign rhythm. The working materials are the campaign enrollment form, the promotional price approval sheet, the inventory lock table and the campaign review sheet. Data arrives from the marketing console, the campaign console, ERP inventory and historical transaction prices. A mistake can produce a loss-making campaign, removal of a resource slot, or a price complaint.
Inventory procurement and supplier records act as the upstream evidence layer in this chain. Purchase orders, supplier quotation sheets, arrival records and settlement statements determine whether campaign inventory can be locked and whether a promotional price has a cost basis. Citation tracing answers one question: can an inventory quantity, supplier quotation or cost basis cited in an enrollment form or approval sheet be traced back to a specific material, a specific version and a specific validity scope?
Materials update weekly or per campaign. The same supplier quotation can therefore correspond to different versions in different campaign rounds, and any check must state the version and the applicable period.
Constraints at the citation tracing step
Retrieval locates candidate citation content. It does not decide whether a material is valid, whether it has been authorized, or whether it can be executed.
The knowledge base search node can associate one or more knowledge bases that share the same vector model, and it outputs citations as an array whose length may be zero. An empty array means the output path still runs; it does not mean the workflow terminated with an error. The citation limit is controlled in tokens. The minimum relevance value is a filter threshold that takes effect only under semantic retrieval or result reranking. Neither of these constitutes a permission boundary or an entity isolation mechanism. Hybrid retrieval merges multiple result paths through RRF; this is a retrieval process, not an authorization switch that can be turned on. Reranking, image retrieval and question optimization each require the corresponding model capability and input conditions, and an optional path shown in a flow diagram does not mean every request executes it.
Citation tracing for inventory procurement and supplier records therefore requires comparing retrieval output against the original materials held by staff, item by item. A retrieval hit does not prove a material is valid, and an empty retrieval does not prove a material does not exist.
Configuration and implementation checks
The settings below concern only the primary task on this page. The values are starting points for a hypothetical exercise, used for comparison against manual verification results. They are not official default values.
| Configuration item | Hypothetical starting point for this exercise | Comparison method and reason |
|---|---|---|
| Associated knowledge bases | Select only the knowledge base holding supplier quotation sheets and purchase orders under one vector model | First check the selected knowledge base identifier against the vector model identifier. Knowledge bases with different vector models cannot be associated together. Exclude and record those that do not match; do not run a cross-model retrieval comparison |
| Search mode | Hybrid retrieval | Supplier names and order numbers suit full-text recall, while cost basis descriptions suit semantic recall. After RRF fusion, compare each returned item against the original text |
| Citation limit | 3000 tokens as the hypothetical starting point for this exercise | Compare against the total length of the material items listed for manual verification. Record anything beyond that length as uncovered; do not claim full recall |
| Minimum relevance | 0.3 as the hypothetical starting point for this exercise | Use only as a candidate filtering starting point. Compare against the items a person judged relevant. Record missing items as unsupported |
| Result reranking | Enable when a rerank model is available | The rerank score is 0-1 and is used for secondary ordering. It is not a basis for authorization or entity filtering |
| Citation handoff | Pass the citation array to the answering step | Quantities, unit prices and supplier names in the answer must be compared word by word against the cited original text. Fluent output is not evidence of source |
Three common errors
- Treating a retrieval hit as proof that a material is valid or authorized. The input is a supplier quotation citation returned by one retrieval. The comparison object is the validity period and signature status in the original quotation sheet. The observable result is that the citation exists while the original text states no effective date. Record this as pending verification rather than judging the quotation valid.
- Treating an empty citation array as a business failure or as proof that a material does not exist. The input is an empty array returned by one retrieval. The comparison object is whether the downstream node continues executing on the empty array. The observable result is that the output path runs while the manual material may still exist. Record the retrieval as not covering the item rather than judging that the supplier issued no quotation.
- Forcing knowledge bases with different vector models into one association and then inferring runtime behavior. The input is two knowledge bases with different vector model identifiers. The comparison object is the set of knowledge base identifiers the node permits to be associated. The observable result is that the two do not meet the shared association condition. Exclude and record them; do not infer missed recall or redundancy.
How to verify the setup
Hypothetical exercise sample. The material originals are: "Supplier A quotation sheet, March 2025: material A unit price 12 yuan, valid until 30 April 2025"; "Purchase order PO-0312: material A, 500 units purchased, expected arrival 10 April 2025"; "Inventory lock table, April campaign: material A, 300 units locked, lock status pending approval". The draft sentence to be judged is: "Material A has 500 units of available inventory and can be enrolled in the April campaign at a cost of 12 yuan."
- Verify the quantity source. The input is the purchase order and inventory lock table originals above. Compare the 500 units in the draft sentence against the order quantity, the locked quantity and the lock status. The observable result is that the order quantity and the locked quantity differ and the lock status is pending approval, so the draft sentence is unsupported.
- Verify the price source and validity period. The input is the quotation sheet original above. Compare the 12 yuan in the draft sentence against the quotation unit price and validity period. The observable result is that the unit price matches while the draft sentence states no validity period, so whether the campaign period falls inside the validity period requires separate verification.
- Verify the empty citation handoff. The input is a paper record of one retrieval returning an empty array. Compare whether the downstream node continues executing and retains a manual verification entry point. The observable result is that the path runs and the gap is recorded; do not conclude from this that the material does not exist or that the business action has ended.