What the documents look like
Franchise filing management works from franchise contracts, operations manuals, filing submissions, and information disclosure documents. The underlying records come from the commercial franchise information management system, enterprise licences, store lists, and contract ledgers. Filing submissions and disclosure documents carry fields such as entity name, signing status, and validity period. In cross-border work, one material may exist in Chinese and in another language at the same time. Retrieval then has to support a judgement about whether two versions point to the same entity, the same period, and the same field.
The category differs from head-office site selection and store opening, which deal with specific premises and engineering. Franchise filing deals with legal disclosure and the commerce department system. Updates are irregular and mainly annual. That cadence does not change how retrieval is configured on this page, so it is not a factor in the settings below.
Constraints at the retrieval step
Multilingual material can match across languages under semantic retrieval, so a Chinese question may hit a block in another language, provided that the material has been ingested and training has completed. Full-text retrieval depends on keyword hits and suits precise queries such as contract numbers, filing numbers, and entity names. Hybrid retrieval merges semantic and full-text results through RRF, and a rerank model can reorder the merged candidates.
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases built on different vector models cannot be associated together. When model identifiers differ, the correct action is to check the identifiers, exclude the non-matching base, and record the actual observation. Cross-model retrieval is not run to compare missed recalls. Reference content is output as an array whose length may be zero, so the output path still runs when nothing is found. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used. It is a filter threshold, not a ranking rule, and it does not perform permission or entity isolation. Retrieval is not authorization. The status of material in an operations handover still depends on manual verification or on a downstream system implementation.
Hypothetical paper exercise sample. The material text to be checked reads: "Party A, XX Chain Management Co., Ltd., signed a Regional Franchise Contract with Party B, signing date March 2023, contract number to be supplied; the filing status column of this contract is marked 'materials submitted, pending review by the competent commerce department'." The draft sentence to be judged reads: "This franchise contract has completed filing and can serve as a basis for handover." The source states only that materials were submitted and review is pending. It does not state that filing was completed. The draft sentence and the source disagree on the same entity, the same period, and the same field, so the draft sentence is unsupported. The handover request "please transfer this contract to operations as filed" likewise lacks evidence of completed filing and is recorded as pending verification.
Configuration and implementation checks
| Configuration item | Setting or check relevant to this page | Relation to the sample |
|---|---|---|
| Associated knowledge bases | Select knowledge bases that use the same vector model; multiple selection is allowed | The filing materials base and the contract ledger base must first have their vector model identifiers checked |
| Search mode | Semantic, full-text, or hybrid retrieval, chosen per test question | Contract numbers use full-text retrieval; cross-language clauses use semantic retrieval |
| Reference limit | Unit is token; adjust after testing with material of different lengths | The hypothetical starting value for this exercise is 2000 tokens, because the sample contract passage plus the filing status note totals roughly 1500 to 1800 tokens; compare against 1000 and 3000 for content completeness |
| Minimum relevance | Takes effect only under semantic retrieval or result reranking; label relevant and irrelevant samples before adjusting | Use the sample source text and the draft sentence as queries and observe whether the source block survives filtering |
| Question optimization | Used for coreference resolution and question expansion; compare the original question with the optimized question | In multilingual follow-up scenarios, check whether the entity has been altered |
| Result reranking | Requires a rerank model enabled under its settings; score range 0-1 | Enable when candidates are numerous, filter by score, and do not presuppose a ranking outcome |
Three common errors
- Reading "materials submitted, pending review" as "filing completed". The input is the sample source text and the draft sentence. The comparison is the value of the filing status field for the same contract. The source value is pending review and the draft value is completed. These are mutually exclusive on the same field, so the draft sentence is unsupported.
- Associating knowledge bases that use different vector models. The input is the model identifier of each base. The comparison is whether the identifiers match. When they differ, exclude the base and record the actual observation. Do not infer a runtime result.
- Treating minimum relevance as a permission or entity filter. The input is a record belonging to another entity. The comparison is whether that record is excluded because of low similarity. A similarity threshold does not perform entity isolation. When authorization evidence is absent, record the gap as pending verification.
How to verify the setup
- Use the sample source text as the query. The input is "materials submitted, pending review by the competent commerce department". Compare whether the returned references contain that source block and its filing status field. The condition is met when the reference content keeps the pending review wording. If it has been rewritten as completed, the condition is not met.
- Use the draft sentence as the query. The input is "this franchise contract has completed filing". Compare the returned references against the sample source text and observe whether the source is hit and shows pending review. If an empty array is returned, record it as not recalled. Do not infer that the rule must return an empty array.
- Check the downstream design with a paper input of an empty result. The input is an explicit empty reference array. Compare whether the downstream node continues to run under the documented condition that length may be zero, and observe whether the output path runs through. Do not claim that a fictional rule makes retrieval always return an empty array.
This page covers only franchise filing management, cross-border material and multilingual retrieval, and knowledge base retrieval and recall. Which materials, operations, or requests sit at which authorization state during an operations handover cannot be settled by retrieval or by a form.