What the documents look like
Franchise filing management handles legal disclosure and the commerce department system. It does not handle specific site engineering. In the store service and internal training setting, knowledge base retrieval works across four document types: franchise contracts, operations manuals, filing materials, and information disclosure documents. The parties involved are the franchisor headquarters, franchise legal staff, and the commerce liaison. Errors in this area can lead to an order to file, franchise contract disputes, or administrative penalties. Data sources include the commercial franchise information management system, enterprise licenses, store lists, and contract ledgers. Update frequency is irregular or annual. The retrieval task is to locate clauses and operating instructions related to store service procedures and training requirements so internal staff can check them.
A retrieval request in this setting typically points at one or two of these document types at a time. A training question about the manual update cycle draws on the operations manual. A question about disclosure obligations draws on the disclosure document and the contract. The same vector model requirement applies when more than one knowledge base is linked.
Constraints at the retrieval step
The knowledge base search node can link one or more knowledge bases that use the same vector model. Knowledge bases with different vector models cannot be linked together; the model identifier is checked first and non-matching bases are excluded. Retrieval output is an array of referenced content, and its length can be zero. The empty-result path still runs through the node.
Search modes are semantic retrieval, full-text retrieval, and hybrid retrieval. Hybrid retrieval merges multiple result paths through RRF. RRF is a fusion process, not a switch that can be turned on. 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. The reference cap is measured in tokens and controls the maximum content referenced per search. Question optimization handles coreference resolution and question expansion, and adds one model call before retrieval. Result reranking uses a ReRank model to reorder candidates and produces a 0-1 score; it depends on the model being available and enabled.
Image retrieval requires image input and the corresponding model capability. It is not part of every request. The overall retrieval flow lists optional paths, and those paths do not all execute on every request.
Hypothetical drill sample: the source text reads "The franchisor shall provide the franchisee with the updated version of the current quarter's operations manual by the tenth day of the first month of each quarter. The franchisee shall complete store personnel training within fifteen working days after receiving the update and retain attendance records. This clause applies to directly operated and franchise stores contracted in the 2024 fiscal year." A draft sentence to be checked reads "After receiving the operations manual update, the franchisee shall complete training record archiving within ten working days." The comparison object is the training completion deadline and the record retention requirement in the source clause.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this drill | Reason and comparison method |
|---|---|---|
| Linked knowledge bases | Franchise contract base, operations manual base | Both use the same vector model; check the model identifier before linking |
| Search mode | Hybrid retrieval | Contract clauses contain numbers and natural language; compare semantic and full-text recall paths |
| Reference cap | Hypothetical starting point of 2000 tokens | A single manual section runs about 1500 to 2500 tokens; compare where truncation falls |
| Minimum relevance | Hypothetical starting point of 0.5 | Label relevant and irrelevant samples first, then compare what remains after filtering |
| Question optimization | Enabled | In multi-turn follow-up scenarios, compare the subject in the original question and the optimized question |
| Result reranking | Enabled as needed | When candidates are numerous, compare rerank scores against the original order |
Three common errors
- Linking knowledge bases that use different vector models. Input: two bases with different vector model identifiers. Compare the model identifiers before the linking action, and observe whether the non-matching base is excluded. Cross-model bases cannot be linked together. Do not run a cross-model retrieval to compare missed recalls.
- Treating minimum relevance as a permission or entity filter. Input: a relevance threshold setting under semantic retrieval. Compare returned content before and after filtering, and observe whether unrelated business records can still be recalled. Minimum relevance only filters low-relevance results and does not exclude unauthorized records.
- Treating a pending-review status as granted authorization. Input: a handover request marked "pending review." Compare the request status field with the authorization field, and observe whether a signed or effective record exists. Pending review does not show that operating permission is absent, and it does not equal completed approval.
How to verify the setup
- For the hypothetical source text and draft sentence, input the draft sentence "After receiving the operations manual update, the franchisee shall complete training record archiving within ten working days." Compare the deadline field in the source, "fifteen working days," with the draft's "ten working days," and observe whether the two match. If they do not match, the draft sentence is not supported by the source.
- For the empty-result path, input a paper retrieval request and expect the referenced content array to be []. Compare whether the node output length is zero, and observe whether the downstream design accepts an empty array. Do not claim that a fictional rule retrieval necessarily returns [].
- For question optimization, input the follow-up question "What is the second point?" Compare whether the optimized question fills in the subject as "training deadline requirements after the operations manual update," and observe whether the subject is altered incorrectly. If the optimized subject does not match the source, record it as pending verification.