What the documents look like
Franchise filing management handles legal disclosure and the commerce department system, not the engineering of a specific store location. The material set comes from the commercial franchise information management system, enterprise licences, store lists and contract ledgers. Day-to-day documents include franchise contracts, operations manuals, filing materials and information disclosure documents. These materials are updated irregularly, with some work proceeding on an annual basis. Operating reviews and exception reviews need to check subject, object, status and validity scope against those documents. The retrieval step only brings candidate content back; it does not perform authorization judgment.
For this page the working scope is franchise filing management, the operating review and exception review task, and knowledge base retrieval and recall. Whether a material, an operation or a request sits at a given authorization state during handover cannot be treated as authorization execution by retrieval or by a form.
Constraints at the retrieval step
A knowledge base search node can associate one or more knowledge bases that use the same vector model. Knowledge bases with different vector models cannot be associated together; check the base identifiers first and exclude those that do not match, rather than running a cross-model search to compare missed recall. Search mode is compared per test question across semantic search, full-text search and hybrid search, without assuming one mode suits all material. The citation cap is measured in tokens, so use actual material of different lengths to test whether content stays complete. The minimum relevance value takes effect only under semantic search or when result reranking is used. Question optimization performs coreference resolution and question expansion, so compare the original question with the optimized question and check whether the subject was altered incorrectly. Hybrid search merges multiple result paths through RRF, and reranking requires a rerank model enabled according to settings. Image retrieval requires the corresponding image input and model capability. Retrieval results do not constitute authorization and do not isolate business entities.
Hypothetical paper exercise sample: a legal staff member at a franchisor headquarters is checking one franchise contract ledger record during an operating exception review. The source material text reads: "Party A, a named catering management company, and Party B, a named franchisee, signed a franchise contract in May 2023, contract number HZ-2023-0512, agreeing that Party A submits filing materials within thirty days of signing. As of January 2024, the ledger registration status is 'filing materials pending submission'." The draft sentence to be judged is: "This contract has completed commerce department filing." The draft sentence and the source material take mutually exclusive values on the filing status field, so the sentence is unsupported.
Configuration and implementation checks
| Configuration item | Hypothetical trial starting point for this page | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the filing materials base and the contract ledger base that use the same vector model | Check base identifiers first; exclude bases with different vector models |
| Search mode | Use full-text search for number-type questions and hybrid search for status-description questions | Compare hit content per test question |
| Citation cap | Set a hypothetical trial starting point of 3000 tokens for this exercise | Compare content completeness using filing material passages of different lengths |
| Minimum relevance | Set a hypothetical trial starting point of 0.4 for this exercise | Label relevant and irrelevant samples first, then compare filtering results |
| Question optimization | Enable for follow-up questions containing references; disable for complete questions | Compare the subject in the original question and the optimized question for consistency |
| Result reranking | Enable when candidates are numerous and a rerank model is available | Compare candidate order and scores before and after reranking |
These rows describe documented settings or explicitly named implementation checks. The numeric starting points are trial values for this exercise only, chosen so that a comparison can be run against labeled samples; they are not official defaults.
Three common errors
- Associating knowledge bases that use different vector models. Input is two base identifiers; compare their vector model identifiers. If they do not match, exclude the base and record the actual observation, without speculating about runtime results.
- Treating minimum relevance as a permission or entity filter. Input is an unrelated business question; compare whether returned content can still include material about other entities, confirming that the parameter only filters low-relevance results.
- Treating a retrieval hit as completed filing. Input is the draft sentence above and the source material text; compare the filing status field. The source says "pending submission" and the draft says "completed", so the draft is unsupported.
How to verify the setup
- Input the source material text and the draft sentence from the hypothetical sample, and compare the filing status field values. If the source reads "pending submission" while the draft sentence states "completed", the draft sentence is judged unsupported and the evidence gap is preserved.
- Input contract number HZ-2023-0512 as a full-text search, and compare whether the returned content includes the passage containing that number. If the returned result is empty, record the empty result and check whether the data has completed training.
- Input a follow-up question containing a reference, and compare the subject in the original question with the subject in the optimized question. If the optimized subject does not match the source material, record the observed subject alteration from question optimization, without concluding a cause of retrieval failure from it.
An empty result can be checked against a downstream design using an explicit paper input of an empty array; this does not assert that a fictional rule search must return an empty array. The overall document flow lists optional paths, and not every request executes all of them: image retrieval requires the corresponding image input and model capability, and reranking requires a rerank model enabled according to settings. Knowledge bases with different vector models cannot be associated together; check identifiers, exclude non-matching bases and record actual observations, without inventing runtime outcomes such as missed recall or redundant returns after a shared association.