What the documents look like
For document comparison in Seller ratings management, this page examines FastGPT tool calls. Reviews, diagnostic statements and rectification material are different evidence. Comparison retains their subjects and stages without inferring completed improvement from a report title. The examples are hypothetical.
Constraints at the tool calls step
Rating-related transaction material can compare service wording, review handling and rectification passages without scores or outcome figures. Proposed input supplies issues and intended correction, not verification. Limit a reading tool to supplied passages, retaining store, order and version. Detailed ratings do not establish every cause, and headings cannot supply diagnostic findings. Inspect system-module and plugin parameters separately without assuming an existing scoring tool.
Configuration and implementation checks
| Transaction-comparison check | Proposed material | Actual evidence |
|---|---|---|
| Order identity | Order named in handling text | Other-order sources not mixed |
| Rectification stage | Plan separate from verification | Not described as resolved |
| Causal evidence | Review statement and diagnosis passage | No direct cause determination |
| Actual invocation | Required passages and real request | Missing sources not guessed |
| Further handling | Separate closure, reply or write | Comparison is not outcome confirmation |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Tool description | Describe purpose, inputs and non-applicable situations. |
Tool parameters | Inspect module and plugin parameters separately. |
Required parameters | The model may not invoke a tool without suitable required inputs. |
Tool connections | Available tools must connect to the calling module. |
Tool-call termination | Termination ends the call without a later AI summary. |
Three common errors
Another order’s handling material needs source correspondence checks. Intended correction described as resolution needs this follow-up evidence. A review statement treated as a service-problem cause needs the actual diagnosis and supporting sources. These concern identity, stage and evidence rather than populated parameters.
How to verify the setup
Inspect actual request, called tool and citations. Repeat with current follow-up material, keeping conclusions bounded. Missing suitable required input may prevent invocation; collect it through interaction and inspect behavior, starting with a small tool set. Termination ends this call without another AI summary. Closure, public replies and writes need authority and real receipts. Differences establish neither improved experience nor responsibility.
| Material context | Scope |
|---|---|
| Document types | Negative Review Handling Records, Experience Score Details, Service Diagnosis Report and Rectification Ledger |
| Candidate sources | Douyin Shop Experience Score, Tmall Service Metrics, JD After-Sales Evaluation and Customer Service System |
| Technical baseline | FastGPT v4.17.2 |