What the documents look like
Franchisee training and opening coaching is a transfer of operating capability rather than a legal filing disclosure. The working materials are training sign-in sheets, opening checklists, SOP manuals and assessment scores. Their wording sets the ceiling for what product and marketing copy can claim.
A sign-in sheet records who attended which course and when. An opening checklist records which materials were in place at a given store on a given date. An SOP manual states operating steps. An assessment score records a numeric result for one learner on one course. When drafting copy, the retrieval step aligns a proposed sentence with these source records and shows whether the promotional wording has supporting material.
These materials update per store and per month, so a retrieved version may lag behind current store conditions. Retrieval configuration needs to let cited content be traced back to a specific document so that a person can check the subject and the period.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Knowledge bases using different vector models cannot be associated together. Check the vector model identifier first, exclude those that do not match, then run retrieval. Retrieval operates on the whole knowledge base; a collection is only a classification layer for data management and does not affect search results.
Semantic retrieval, full-text retrieval and hybrid retrieval are compared separately against a test question. No single mode is presumed suitable for all materials. Hybrid retrieval merges multiple result paths through RRF; RRF is a fusion process and is not an independent switch. Result reranking requires a ReRank model and must be enabled in settings. Image retrieval requires image input and the corresponding model capability. Optional paths shown in the overall flow diagram do not mean every request executes all of them.
The minimum relevance value is a filter threshold and takes effect only in semantic retrieval or when result reranking is used. It does not perform permission or entity isolation. The citation limit is measured in tokens. An empty result is output as an array whose length can be 0.
Configuration and implementation checks
Hypothetical paper exercise sample. Source text A: "March 2025 East China franchisee cohort 2 cashier training sign-in sheet: learner Li completed the cashier practical course on 10 March, assessment score 82." Source text B: "March 2025 East China opening checklist: at a franchise store on 12 March, cashier counter materials were complete and the cashier process followed chapter 4 of the SOP manual." Proposed draft sentence under review: "This franchisee has completed cashier training and operates the cashier process proficiently." This sample is a paper exercise and does not refer to a real customer.
| Configuration item | Value or check method on this page | Relation to the sample |
|---|---|---|
| Associated knowledge bases | Associate only the training and SOP bases that use the same vector model | Keeps the sign-in sheet and checklist inside one retrieval scope |
| Search mode | Use the draft sentence as the test question and compare semantic and full-text results separately | Shows whether "cashier training" and "cashier process" hit the matching documents |
| Citation limit | For this paper exercise the starting value is set at 512 tokens, compared against the combined length of the sign-in sheet and checklist text | The starting value is a setting for this paper exercise, not an official default |
| Minimum relevance | Label relevant and irrelevant samples first, then compare thresholds | Takes effect only in semantic retrieval or result reranking |
| Question optimization | Compare the original draft sentence with the optimized question and check whether the subject was altered | Used for coreference resolution and question expansion |
| Result reranking | Compare scores when a ReRank model exists and reranking is enabled in settings | When not enabled, this path does not participate |
Three common errors
- Treating the assessment score of 82 as evidence of proficient operation. The score is a numeric record in one field; proficient operation is a capability judgment. The subject and the field differ, and the materials do not state a proficiency level, so the claim remains unsupported.
- Treating a retrieval hit as business authorization. Returned cited content only indicates textual similarity. It does not mean the franchisee has been approved to open or that promotional wording has been approved.
- Forcing knowledge bases that use different vector models into one association. Knowledge bases with different vector models cannot be associated together. Check the identifier, exclude those that do not match, and record the actual observation without inferring runtime results.
How to verify the setup
- Enter the source texts and the draft sentence above, then compare whether the retrieved citations cover both the sign-in sheet and the checklist. If the citations contain only the sign-in sheet, the part about the cashier process following the SOP is unsupported and the gap must be recorded.
- Enter the draft sentence and compare the cited content before and after the minimum relevance filter. If the filtered citations form an empty array, the downstream design must accept an empty result. Do not claim that a rule-based search necessarily returns an empty array.
- Enter the draft sentence and compare the question text before and after question optimization. If the subject changes from "a franchise store" to another subject after optimization, the subject was altered incorrectly and that retrieval result is not used for copy review.