What the documents look like
For after-sales questions in Brand repair services, this page examines FastGPT retrieval. Retrieve inspected items and pending scope without turning a partial check into restored use. The examples are hypothetical.
Constraints at the retrieval step
A proposed brand-repair case completes an indicator-light check, leaves startup testing pending and asks whether every function is restored. Partial checking and unresolved testing can coexist without a whole-device result. Retrieve indicator and startup subjects with their states, comparing the complete question with literal test names. A completed check cannot cover every function, while pending startup testing does not establish a fault. The question requires applicable inspection evidence; changing wording does not create it or determine the actual condition of a device from these hypothetical service records.
Configuration and implementation checks
| Material check | Conditions in this test | Implementation evidence |
|---|---|---|
| Inspection scope | Indicator item | Do not expand to all functions. |
| Pending test | Startup | Retain in reference. |
| Optimization scope | Still all functions | Do not narrow task. |
| Clipped positive state | Complete needs its item | Inspect actual text. |
| Added result | Device, item and version | Update matching scope only. |
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, local completion may establish full restoration; inspect candidate test and conclusion scope. Second, optimization may replace all functions with whether the light was checked; retain question scope rather than narrowing the task before saying complete. Third, reference trimming may remove pending testing; compare the original with text actually delivered. These concern evidence scope, task scope and reference completeness. Scores, record counts and longer answers cannot supply a whole-device state when the available source only covers one finished check and another unresolved item.
How to verify the setup
Ask which check finished, which remains pending and what full restoration would require, expanding actual quotations. Compare semantic, full-text and hybrid modes using the same questions, retaining item names with finished or pending states under token limits. Added inspection needs device, item and version matched, updating only its supported scope. This exercise diagnoses nothing, prescribes no repair and decides no warranty responsibility. It guarantees no real restored function. Missing outcomes remain unresolved rather than a usable device inferred from a relevant completed-check passage.
| Material context | Scope |
|---|---|
| Document types | Warranty Card, Repair Work Order, Inspection Report and Exchange Approval |
| Candidate sources | After-Sales System, Store Orders, Repair Service Point Feedback and National Three Guarantees Catalog |
| Technical baseline | FastGPT v4.17.2 |