What the documents look like
For product questions in Franchise store supervision, this page examines FastGPT hybrid retrieval. Compare the material received with the sources used in the answer. The examples are hypothetical.
Constraints at the hybrid retrieval step
“Second shelf” and “promotional label attached” in the proposed store-visit form record display information, without describing product performance. Use product A and both recorded details to mark required candidates, then inspect whether reranking can support separate answers about location and label status. Retain the product association. Another item’s display standard cannot become an event recorded in this visit. This is a hypothetical text test, without verification of an actual store display or promotion.
Configuration and implementation checks
| Item or implementation check | Input or applicable condition | Traceable check |
|---|---|---|
| Fused candidates | Second-shelf location and promotional-label record | Check association of both details with product A. |
| Complete question | Ten units recorded versus twelve prescribed | Keep recorded and prescribed values distinct. |
| Minimum relevance | Inspect filtering under semantic or reranking conditions | A standard-value hit does not verify the store quantity. |
| Citation token limit | Display location, quantity and source role | Do not turn a standard into a visit record. |
| Missing-source retest | No new-item designation in the visit form | A promotional label does not establish new-item status. |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Hybrid retrieval | Vector and full-text results use RRF fusion. |
| Reranking | Reranking reorders candidates using the full question. |
Minimum relevance | Filtering applies to semantic search or reranking. |
Citation limit | Citation content is limited by tokens. |
Question optimization | Compare subjects and restrictions after optimization. |
Three common errors
Ten units in the visit report and twelve in the display standard describe a recorded value and a prescribed value. Preserve those roles instead of selecting one quantity as the unified answer. Inspect candidates for both product-and-quantity statements, then check the applicable standard version and scope. The two numbers identify a review point; whether there is an operational issue requires the corresponding business confirmation, not a conclusion supplied by reranking. Preserve the roles of observed visit wording and prescribed display wording with their citations. A standard value must not become a measured store value. Whether rectification is needed depends on actual applicable requirements, beyond comparing two numbers in candidates.
How to verify the setup
The existing report does not explicitly identify the item as new. Shelf location and a promotional label cannot supply that conclusion. Use the new-product question to inspect whether the answer infers launch status from promotion wording. After adding an explicit designation or corresponding product material, inspect its path through candidates, reranking and citations. Check minimum relevance under the actual semantic or reranking conditions, without treating the score as evidence of new-product status.
| Material context | Scope |
|---|---|
| Document types | Store Visit Report, Franchise Contract, Rectification Notice and Display Standards |
| Candidate sources | Supervisor APP, POS Sales, Franchise System and Store Photos |
| Technical baseline | FastGPT v4.17.2 |