What the documents look like
Franchise fee and deposit settlement relies on four document types: the payment schedule inside the franchise contract, the deposit ledger, deduction notices, and franchise fee invoices. Together they determine whether a fee is receivable, already collected, or refundable. This work concerns money and breach deductions. It sits apart from franchise filing administration, so filing documents do not serve as settlement evidence during retrieval.
The brief records updates at monthly intervals and at each deduction. One franchisee may therefore have several ledger rows across different months. Retrieved references must distinguish month and deduction reason. If they do not, a draft sentence can draw on amounts or conditions from the wrong period. During copy drafting, the writer checks whether a proposed sentence about an amount, a deadline, or a refund condition is supported by these documents. The retrieved reference content is the basis for that check.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that share the same vector model. Knowledge bases built on different vector models cannot be associated together. The operator checks the vector model identifier first, excludes non-matching bases, and records the association scope actually observed.
Semantic retrieval compares vector distance. Full-text retrieval depends on keyword hits. Hybrid retrieval merges both paths through RRF. Result reranking applies a ReRank model for a second ordering pass. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used. It is a filter threshold. It does not act as a permission filter or an entity isolation mechanism. The reference limit controls how much content each search may cite, measured in tokens. Question optimization performs coreference resolution and question expansion, and it adds one model call before the actual retrieval.
Hypothetical paper exercise. Source text one: "Party A shall pay a franchise fee of RMB 50,000 and a deposit of RMB 30,000 within ten days of signing. The deposit is refundable upon contract expiry when no unsettled deduction remains." Source text two: "The March 2024 deduction notice states: due to non-compliant in-store display, RMB 2,000 is deducted from the deposit; the deduction reason is display non-compliance." The draft sentence under review: "The franchisee deposit is fully refundable in all circumstances, and the franchise fee may be paid in installments." This draft sentence receives no support from source texts one and two on the refund condition or the payment method. It remains unverified and cannot go into external copy as written.
Configuration and implementation checks
The table below lists settings or implementation checks that the official documents describe, tied to the sample on this page. The values are hypothetical starting points for this exercise, used for comparison against the sample. They are not official defaults.
| Configuration item | Hypothetical starting point | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only bases holding the franchise contract and ledger under one vector model | Check the vector model identifier first; exclude bases on other models |
| Search mode | Hybrid retrieval | Amounts and identifiers need keyword hits; refund conditions need semantic matching |
| Reference limit | Set to 2000 tokens for this exercise | Compare against sample source length; confirm the refund condition sentence enters the reference intact |
| Minimum relevance | Set to 0.4 for this exercise | Test the draft sentence and the source sentences separately; observe whether unrelated fragments are filtered |
| Question optimization | Enabled | Compare the original question with the optimized question; check whether the subject was altered |
| Result reranking | ReRank enabled | With many candidates, compare scores before and after reranking; confirm the refund condition sentence position |
Three common errors
- Treating minimum relevance as a permission filter. Minimum relevance filters low-relevance results only. It does not exclude unrelated businesses or unauthorized records. Retrieval itself is not authorization; business verification and approval require downstream implementation.
- Describing RRF as a switch that can be turned on. RRF is the merge process for multiple result paths inside hybrid retrieval, not a standalone configuration item.
- Treating a pending review state as missing operation permission. Pending review means the process is incomplete. Signing, reaching cooperation, and contract effectiveness are three distinct facts.
How to verify the setup
- Input source text one and the draft sentence "the deposit is fully refundable in all circumstances." Compare the refund condition "upon contract expiry when no unsettled deduction remains" in the source with the absolute wording of the draft. If the retrieved reference contains that condition sentence in full and the draft sentence is marked as unsupported, the expectation is met. If the reference omits that condition sentence, record it as incomplete recall.
- Input source text two and the query "deposit deduction reason." Compare whether the retrieved content contains both "display non-compliance" and the deduction amount. If both the reason and the amount appear, the expectation is met. If only the amount appears and the reason is missing, record it as an incomplete field.
- Input an empty-result continuation case, using an explicit paper input of an empty array to check the downstream design. Compare whether the downstream node still runs its output path when the reference length is 0. If the downstream continues with an empty reference, the expectation is met. Do not claim that a fictional rule search necessarily returns an empty array.
All three items use the hypothetical sample as input and were not actually executed. Knowledge bases on different vector models cannot be associated together; the check confirms the identifier and excludes non-matching bases, and does not run a cross-model retrieval comparison.