What the documents look like
For after-sales questions in Store price and display checks, this page examines FastGPT multi-turn prompts. Carry price-label periods, product scope and attributed store replies into follow-up service questions. The examples are hypothetical.
Constraints at the multi-turn prompts step
A hypothetical conversation starts with an audit note saying an old price label awaits removal and its promotion period has ended. A customer then asks whether a recent purchase follows the displayed terms. The note supplies neither this transaction’s receipt nor a store decision. The prompt should request purchase time, product and label correspondence before organizing what can be explained. A recent purchase must not refer automatically to another customer’s historical order. A label still displayed does not establish continuing promotional applicability. This exercise examines text and references without deciding transaction rights.
Configuration and implementation checks
| Audit follow-up | Exercise subject | Inspection |
|---|---|---|
| Period | Old label period ended | Actual purchase time |
| Product | Recent purchase identity | Corresponding label |
| Reference | Earlier order may concern another store | Original versus optimized question |
| Reply | Store proposes follow-up | Customer response separately |
| Gap | Missing receipt versus omitted input | Different remedial paths |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Prompt and task | Define sources, the task and evidence-gap handling. |
Conversation history | Inspect the history actually submitted. |
Question optimization | Review reference resolution and question expansion. |
Model and length | Inspect context and reply length separately. |
Reply format | json_schema needs its format and parsing validation. |
Three common errors
- If the next request omits the ended-period sentence but answers that the offer still applies, retain the request and history to locate the lost period restriction.
- If question optimization substitutes an earlier order from another store, compare the original question, optimized question and order identity. Trace the changed store and product.
- If a proposed store follow-up becomes customer acceptance, inspect its author, actual reply and customer response. A planned action supplies no mutual outcome.
How to verify the setup
Compare the same follow-up with empty history, the audit note alone and the note plus this purchase’s evidence. Record remaining gaps. Inspect product and transaction time when the receipt arrives; upload time is a separate event. Attribute later store statements to their speaker. A source sentence obtained but omitted calls for context investigation; an unobtained receipt calls for collection. More history or a longer answer cannot create customer acceptance. This exercise changes no price, issues no refund and promises no remedy.
| Material context | Scope |
|---|---|
| Document types | Store Inspection Photos, Price Tag List, Display Standards and Rectification Notice |
| Candidate sources | Supervisor System, POS Price List, Store Cameras and Headquarters Promotion Schedule |
| Technical baseline | FastGPT v4.17.2 |