What the documents look like
For product questions in B2B marketplaces, this page examines FastGPT retrieval. Compare the material received with the sources used in the answer. The examples are hypothetical.
Constraints at the retrieval step
A new B2B quotation test can state, “The valve body uses a standard interface; sealing material requires separate confirmation for the intended use.” Interface form and material suitability need different evidence. Retrieve the interface wording, then compare semantic, full-text and hybrid search with the complete use-qualified question. Returning the product name does not answer the material question. Citations must retain “separate confirmation,” without extending one description into established suitability for every application.
Configuration and implementation checks
| Documented item or implementation check | Material and condition | Evidence action |
|---|---|---|
| Source scope (implementation check) | Quotation and sample record remain distinct sources | A demonstration record does not replace material information. |
| Search modes: semantic, full-text, hybrid | Literal interface wording and the use-qualified question | Compare originals returned by the three modes. |
| Citation token limit | Keep the use restriction beside the interface description | Inspect any omitted separate-confirmation condition. |
| Minimum relevance | Inspect filtering under semantic search or reranking | A high score does not establish application suitability. |
| Question optimization | Follow-up asking whether this model can be used | Check that optimization adds no permission or use assumption. |
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
A separately proposed sample record says only that the interface style was demonstrated. Use it to inspect whether nearby terminology displaces the actual use restriction. Retain both source identities and establish which candidates come from the quotation or demonstration record. Check whether the answer converts a demonstration into verified suitability. Citation limits are token-based, beyond snippet counts. If the use condition is trimmed, record the omitted location and repeat the same question against the original.
How to verify the setup
Without the sealing-material explanation, the supplied evidence consists of an interface description and a sample record. Question optimization may resolve references but cannot provide the absent material support, including an invented application permission for “can this model be used?” After obtaining the corresponding explanation, repeat the question and inspect entry of the new original into candidates and citations. Check minimum relevance under semantic or reranking conditions, without treating a score as a decision on actual suitability.
| Material context | Scope |
|---|---|
| Document types | Business Certification Documents, Inquiry and Quotation Sheet, Electronic Contract and Platform Settlement Statement |
| Candidate sources | B2B Platform, National Enterprise Credit Information Publicity System, Payment Transaction Records and Logistics Tracking |
| Technical baseline | FastGPT v4.17.2 |