What the documents look like
For after-sales questions in Brand-owned store operations, this page examines FastGPT retrieval. Check store aliases, collection notices and common opening notes against actual location identity. The examples are hypothetical.
Constraints at the retrieval step
A proposed company-store order names South Street for collection, a notice says South Store and a general note opens both locations at weekends. The customer asks whether the names identify one place. No alias mapping or address correspondence is supplied. Retrieve both complete names with the identity question; a shared word or brand does not establish that the names refer to the same location. The opening note explains no naming relationship. A notice describes collection at its named store without automatically determining where the current order belongs or proving the customer should attend that location.
Configuration and implementation checks
| Material check | Conditions in this test | Implementation evidence |
|---|---|---|
| Order location | South Street | Retain complete name. |
| Notice location | South Store | Do not expand automatically. |
| Evidence content | Shared opening hours do not establish a store-alias relationship | Inspect actual mapping text. |
| Reference location | Collection needs store name | Inspect clipping. |
| Added mapping | Issuer and store scope | Exclude other-city names. |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Associated knowledge bases | Knowledge bases associated with one search node must use the same embedding model. |
Search mode | Semantic, full-text and hybrid modes are available. |
Citation limit | Citation limits are measured in tokens. |
Minimum relevance | Range 0–1; active for semantic search or reranking. |
Question optimization | Resolves references and expands questions. |
Three common errors
First, common opening hours may become alias evidence; inspect whether candidates actually provide name or address correspondence. Second, optimization may expand South Store directly into South Street; compare entities instead of assuming the relationship before searching. Third, references may keep collection but lose the store name; inspect retained location wording rather than guessing from the order. These concern evidence content, entity rewriting and reference subject. High scores or more same-brand records cannot supply an absent mapping between the two names in the proposed packet.
How to verify the setup
Ask which store the order names, which the notice names and what the opening note establishes, expanding actual references. Compare semantic, full-text and hybrid results with both names unresolved as distinct entities. Token budgets need complete location names. Added alias evidence requires issuer and store scope checked without borrowing a same-named store from another city. This exercise directs no real customer, changes no order and sends no notice. Unverified naming remains pending rather than an incorrect notice or identical locations inferred from similar search wording.
| Material context | Scope |
|---|---|
| Document types | Daily Sales Report, Staff Schedule, Display Photos and Inventory Transfer Order |
| Candidate sources | POS System, Member CRM, ERP Inventory and Store Inspection |
| Technical baseline | FastGPT v4.17.2 |