What the documents look like
For product questions in Convenience fresh-food orders, 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
The fresh-food example supplies a suggested storage range of zero to four degrees Celsius for one sandwich in a proposed order attachment. Mark the product, range and the word “suggested” as the reading target before inspecting candidate completeness. Do not transfer another item’s temperature wording to this sandwich. This is a check against hypothetical source text, without providing real food-storage advice or establishing the appropriate conditions for an actual product.
Configuration and implementation checks
| Item or implementation check | Input or applicable condition | Traceable check |
|---|---|---|
| Fused candidates | Sandwich and suggested zero-to-four-degree wording | Retain the item and the suggestion qualification. |
| Complete question | Suggested range versus recorded range for the beverage | Distinguish the zero-to-four suggestion from the zero-to-six record. |
| Minimum relevance | Inspect during semantic retrieval or reranking | Filtering does not determine appropriate real storage conditions. |
| Citation token limit | Temperature values together with their context | Do not lose the record’s role while retaining the range. |
| Missing-source retest | No allergen information for the dessert | Do not infer ingredients from a name or another dessert. |
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
The beverage order gives a suggested range of zero to four, while the temperature record states zero to six. These are different source roles. Inspect both candidates, establish product and period correspondence, then check whether reranking encourages a recorded range to be described as the recommended standard. Examine token trimming for lost distinctions between a suggestion and a record. Actual conformity needs the applicable evidence and complete records, beyond the ranking result. Inspect the situation described by each passage as well as its range. The available wording supplies no complete temperature history, so output must not invent continuous conformity or continuous deviation. Preserve the limited evidence scope alongside the two source statements.
How to verify the setup
No allergen information has been supplied for the dessert. Its name or ordering temperature cannot replace ingredient evidence. Inspect whether high-ranked candidates concern another dessert or the answer makes unsupported claims about the presence or absence of an ingredient. After obtaining the actual product material, repeat the question and trace the new citation, checking filtering and ordering separately. Reranking cannot generate absent allergen evidence, and relevance does not confirm a food attribute.
| Material context | Scope |
|---|---|
| Document types | Fresh Food Order Form, Spoilage Report, Near-Expiry Processing Sheet and Temperature Records |
| Candidate sources | POS Sales, Headquarters Ordering System, Store Inventory and Cold Chain Temperature Control Records |
| Technical baseline | FastGPT v4.17.2 |