Franchise fee settlement: retrieval for operating reviews

Review retrieval for operating reviews in Franchise fee settlement. Check document scope, failure paths and source evidence through controlled tests.

What the documents look like

Franchise fee and deposit settlement work revolves around money movement and breach deductions. It does not handle government filing. The materials that matter for this page are franchise fee invoices, deposit ledgers, deduction notices, and contract payment schedules. They originate from the finance system, franchise contracts, bank statements, and store violation records. Operating reports and exception reviews retrieve and cite these materials to reconcile payment terms, confirm deduction reasons, and check invoicing status. Retrieval accuracy directly affects how well a dispute over funds or a deposit refund disagreement can be reviewed. Materials update monthly and on each deduction, so the same store and the same period may draw fields from different documents.

A franchise contract payment schedule might state that a party shall pay a franchise fee of 80,000 yuan and a deposit of 50,000 yuan by 31 March 2025, and that overdue deductions follow the store violation records. A deposit ledger might show a deposit balance of 50,000 yuan for that store in March 2025. A draft sentence under review might read that the store's March 2025 franchise fee has been settled. Checking it requires confirming that the invoice, bank statement, and contract payment schedule share the same subject, the same period, and the same field. If the invoice is missing or the period does not match, the draft sentence is not supported.

Constraints at the retrieval step

A knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases with different vector models cannot be associated together. The search mode can be semantic retrieval, full-text retrieval, or hybrid retrieval. Hybrid retrieval merges multiple result paths through RRF, and ReRank performs a second ordering of candidate text. The reference cap is measured in tokens. The minimum relevance threshold takes effect only under semantic retrieval or when result reordering is used. Question optimization handles reference resolution and question expansion, and it adds one model call before the actual retrieval. Reference content is output as an array, and its length can be zero, meaning the downstream chain still runs even when nothing is found.

Retrieval does not perform permission or entity isolation. The minimum relevance threshold is a filter value, not a business authorization decision. A relevance score is not a category, entity, or permission filter, and it does not exclude unrelated businesses or unauthorized records. Retrieval is not authorization. The same-model requirement for associated knowledge bases does not extend to business registration sites or other external systems.

The document flow diagram lists optional paths that do not all execute on every request. Image retrieval requires a corresponding image input and model capability. Reordering requires a model and must be enabled according to settings. Cross-model bases cannot be associated, so the check is to verify the model identifier, exclude bases that do not match, and record what is observed rather than running a cross-model search to compare missed recalls.

Configuration and implementation checks

Configuration itemHypothetical starting point for this exerciseReason and comparison method
Associated knowledge basesFranchise contract base, deposit ledger base, deduction notice baseThe three bases use the same vector model; verify the model identifier before associating
Search modeHybrid retrievalContract clauses match numbers through full-text retrieval; deduction reasons match descriptions through semantic retrieval
Reference cap3000 tokensContract payment schedule paragraphs are long; compare content completeness against 1500 tokens
Minimum relevance0.4Label relevant and irrelevant samples first, then compare filtering against 0.3 and 0.5
Question optimizationEnabledMulti-turn follow-up such as "what about this deduction" needs reference resolution; compare against disabled to check whether the subject is changed incorrectly
ReRankEnabledSecond ordering when candidates are numerous; compare ordering stability against disabled

Three common errors

  1. Treating the minimum relevance threshold as a permission filter. Input a query for store A's deposit balance, compare whether returned content includes store B records, and observe whether cross-store results appear. A relevance score does not perform entity isolation.
  2. Associating knowledge bases that use different vector models. Input two bases with different model identifiers, compare whether the association is permitted, and observe whether a message about the inability to associate appears. The documentation states that bases with the same vector model can be associated.
  3. Writing RRF as a switch. Input a hybrid retrieval configuration, compare whether an RRF switch item exists, and observe whether the configuration interface shows only search mode options. RRF is a fusion process, not a standalone switch.

How to verify the setup

  1. Input the contract payment schedule text from the hypothetical sample and the draft sentence that the store's March 2025 franchise fee has been settled. Compare the subject, period, and field across the invoice, bank statement, and contract payment schedule, and observe whether the three agree. If the invoice is missing or the period does not match, the draft sentence is not supported.
  2. Input a retrieval request for store A's deposit balance. Compare whether store B's deposit records appear in the returned references, and observe whether only store A content is returned. If cross-store results appear, the minimum relevance threshold does not perform isolation.
  3. Input a clearly labeled paper empty-result scenario. Compare whether the reference content array is empty, and observe whether the downstream chain continues to execute. An empty-result chain running through does not mean a retrieval rule necessarily returns an empty array.

A pending review does not demonstrate a lack of operating permission. Signing, reaching cooperation, and contract effectiveness are three distinct facts. A quotation validity period does not indicate a supply period. Do not add authorization or effectiveness clauses to obtain a definite answer; record the evidence gap instead. Materials needed for verification must sit within the target retrieval scope to which access is held or within a clearly identified manual input; when missing, record the gap.