What the documents look like
Supermarket supplier entry settlement is built mainly on purchase-sale or consignment payables. A supplier does not necessarily operate a staffed counter on site. The everyday document set is made up of the purchase contract, the delivery acceptance note, the rebate agreement and the reconciliation statement. These materials come from ERP procurement, the supplier portal, POS sales and the finance system, and they update weekly or monthly.
Citation tracing at the inventory procurement and supplier records step answers a narrow question: which material, which version and which passage a given piece of cited content came from. Document differences feed directly into reconciliation. When the accepted quantity on a delivery acceptance note and the quantity on a reconciliation statement disagree, the only reliable route is back to the original instruments, item by item. A system summary figure is not a substitute for that comparison.
Two characteristics of this category shape the task. First, settlement here is accounts-based rather than counter-based, so the evidence chain runs through contracts, notes and statements rather than through on-site operating records. Second, the documents arrive from several systems at different cadences, so a citation that looks internally consistent may still rest on an outdated version of one instrument.
Constraints at the citation tracing step
The knowledge base search node outputs cited content as an array, and the length may be zero. This means the output path still runs even when nothing was retrieved. The citation ceiling controls how much content a single search may cite, measured in tokens. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used; it is a filter threshold and does not perform permission or entity isolation.
Multiple knowledge bases can be associated together only when they share the same vector model. Cross-model association is not available, so the identifiers must be checked and non-conforming bases excluded before association. After retrieval output is passed to the answering step, the answer and the cited source text must be checked line by line. A thinking process or a fluent answer does not replace source evidence.
Hypothetical paper exercise sample
The following materials are used for this paper check. They do not refer to any real customer or real contract.
Material one, hypothetical purchase contract excerpt: "Party A is a supermarket. Party B is a food supplier. The contract period runs from 1 January 2025 to 31 December 2025. The entry fee is deducted in one instalment from the first payment."
Material two, hypothetical delivery acceptance note excerpt: "Acceptance date 5 March 2025. Item category biscuits. Accepted quantity 120 cases. Acceptance conclusion qualified."
Material three, hypothetical reconciliation statement excerpt: "Reconciliation period March 2025. Purchase quantity 120 cases. Return quantity 5 cases. Amount payable calculated on net quantity."
Draft sentence to be assessed: "The supplier's accepted quantity for March 2025 was 120 cases and the reconciled quantity was 120 cases, so the two agree."
Configuration and implementation checks
The table below covers this page's sample and lists officially described settings or implementation checks relevant to this feature and primary task.
| Configuration item | Hypothetical starting point for this exercise | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Select the purchase contract base and the acceptance note base that share one vector model | The official description allows joint association only for identical vector models; check identifiers first, then associate |
| Citation ceiling | Assume 3000 tokens for this exercise | A starting point for this exercise, used to observe whether cited content is truncated, compared against the combined length of the three sample materials |
| Minimum relevance | Assume 0.3 for this exercise | A starting point for this exercise, effective only under semantic retrieval or reranking, compared against cited item counts at other values |
| Search mode | Hybrid retrieval | Identifier and date queries suit full-text retrieval; natural language questions suit semantic retrieval |
| Result reranking | Enabled as needed | Reranking requires a model and must be enabled by configuration; it does not run on every request |
| Citation output handoff | Check whether the output array is passed to the answering step | Output length may be zero; an empty array does not mean the workflow terminated with an error |
Three common errors
- Treating a retrieval score as a permission filter. The minimum relevance value is a filter threshold and does not exclude unrelated businesses or unauthorised records. Check: input the three sample materials and observe whether the cited array contains all three; if one is missing, compare its source text against the query terms, record the absence as a fact, and do not infer a permission block.
- Treating a pending state as an approval. Material two records an acceptance conclusion of qualified, and material three covers the March 2025 reconciliation period. The two share a subject and a period but carry different fields. Check: verify whether the draft sentence merges the accepted quantity and the reconciled quantity into one field, and mark a field mismatch if it does.
- Forcing association of knowledge bases with different vector models. The official description states that only identical vector models may be associated together. Check: verify the vector model identifier of each selected knowledge base, exclude non-conforming ones, and record the actual observation. Do not run a cross-model search to compare missed recalls.
How to verify the setup
The three checks below use the hypothetical source text and draft sentence given above, and state the input, the comparison and the observation condition for each. All three are paper checks and were not actually executed.
- Input the material one source text and the draft sentence, and compare whether the contract period from 1 January 2025 to 31 December 2025 covers the March 2025 period stated in the draft sentence. Observation condition: if the period stated in the draft sentence falls inside the contract period, the period scope is consistent; if the stated period falls outside the contract period, mark a period inconsistency.
- Input the material two source text and the draft sentence, and compare the accepted quantity of 120 cases with the 120 cases stated in the draft sentence. Observation condition: if the values match and the subject, date and item category also match, the accepted quantity field is consistent; if the draft sentence does not state the acceptance date or the item category, mark an evidence gap and leave it pending verification.
- Input the material three source text and the draft sentence, and compare the reconciled quantity with the reconciled quantity stated in the draft sentence. Observation condition: material three records a purchase quantity of 120 cases and a return quantity of 5 cases; if the draft sentence treats the purchase quantity as directly equal to the reconciled quantity, mark a field mismatch; if the draft sentence does not mention the return quantity, mark it as unsupported and leave it pending verification.
For the empty-result handoff, a clearly labelled paper input of an empty array can be used to check the downstream design and observe whether the answering step runs through normally. This does not assert that a fictional rule search will necessarily return an empty array.