What the documents look like
Site selection and store opening happen before a store begins trading. The core materials are lease and engineering delivery documents, not trading records. The main data sources named in the brief are trade-area foot traffic, map POI, POS historical store types, and landlord materials. Updates are irregular and per store. The daily documents are a site evaluation form, a letter of intent for lease, a renovation budget, and an opening schedule. The three most time-consuming steps are calculating the rent-to-sales ratio, negotiating lease terms, and coordinating renovation milestones.
Citation tracing for after-sales questions concerns a narrow question: when an answer is produced, which material, which version, and which sentence does it point to. The materials carry different stages. A site evaluation form records an assessment. A letter of intent records an intention, not a lease. A renovation budget records a cost plan. An opening schedule records planned dates. These stages do not collapse into one another. A signed letter of intent is not a signed lease, and a signed lease is not proof that a deposit is refundable. The tracing task is to keep the cited source and its stated stage separate from any downstream business action.
Constraints at the citation tracing step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. The output is an array of cited content, and its length can be zero. A zero-length array still completes the output path. Retrieval does not carry authorization duties. The minimum relevance value is a filter threshold and does not act as a permission or entity isolation control. Mixed retrieval combines multiple result paths through RRF, which is a fusion process, not a switch. Reranking requires a rerank model and must be enabled in configuration. Image retrieval requires image input and model capability. Optional paths shown in the overall retrieval diagram do not all execute on every request. Knowledge bases with different vector models cannot be associated together; the identifiers must be checked first and non-matching ones excluded.
Hypothetical paper exercise sample. The material text reads: "Party A and Party B signed the Letter of Intent for Lease on 1 March 2025, covering Unit B in Trade Area A, with an intent deposit of RMB 50,000, valid until 31 March 2025." The draft sentence to be checked reads: "The lease contract for this unit took effect in March 2025." The source text records only the signing of a letter of intent and its validity period. It records no contract signing and no effective date. The draft sentence is therefore unsupported by the cited source. A second handover request reads: "Please process the deposit refund based on the above letter of intent." The intent deposit and the deposit are different fields, and a refund is a downstream business action. Retrieval and a form do not constitute authorization to execute it.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison object |
|---|---|---|
| Associated knowledge bases | One base for the site evaluation form, one for the letter of intent | Compare the vector model identifiers of both bases before associating |
| Search mode | Mixed retrieval | Compared with semantic retrieval, numbered materials rely more on full-text matching |
| Citation limit | To be set according to material length | The document measures by tokens, not by item count |
| Minimum relevance | Left unchanged for now | Compare against the displayed value before deciding |
| Result reranking | Enabled when a rerank model is available | Compare candidate ordering with and without reranking |
| Question optimization | Enabled for multi-turn follow-up scenarios | A single clear question can leave it off to reduce latency |
Three common errors
- Treating retrieval output as authorization. The search node outputs a citation array. A non-empty array only shows that recalled content exists. It does not show that a deposit is refundable or that a contract has taken effect. The input is the citation array, the comparison object is the state recorded in the original material, and the observable result is whether the two agree.
- Treating a signed letter of intent as an effective contract. The sample source text records only the signing of the letter of intent and its validity period. The draft sentence claims the contract has taken effect. The input is the draft sentence, the comparison object is the fact recorded in the source, and the observable result is that the claim is unsupported.
- Forcing association of knowledge bases with different vector models. The input is the vector model identifier of each base, the comparison object is whether the two identifiers match, and the observable result is that a mismatch requires exclusion rather than a cross-model retrieval comparison.
How to verify the setup
- Test citation tracing with the sample source text. Input the material text and the draft sentence, compare whether the answer's citation points to that source, and observe whether the cited content can be matched sentence by sentence to the source and whether the letter of intent signing is presented as an effective contract.
- Test downstream handling of an empty citation array. Input a clearly written array [] on paper, compare whether the downstream node handles it as zero length, and observe whether the output path completes without terminating on error.
- Test material timeliness with a historical version. Input the current version and a historical version of the letter of intent, compare which version the answer cites, and observe whether the cited version is identifiable. Whether it is adopted still requires review against the effective version.
These checks describe expected observations. They do not report that any test has been run, that any deployment has occurred, or that any approval or refund has taken place.