What the documents look like
Regional dealer stock replenishment runs on day-to-day rolling orders rather than seasonal centralized ordering. The materials in scope for this page are replenishment requests, inventory turnover sheets, outbound notes and repayment plans, updated weekly or monthly. Contract and transaction material comparison checks whether the subject, object, status and valid scope line up across those materials.
A replenishment request states the requesting subject and the replenishment object. An inventory turnover sheet states the turnover period and stock quantity. An outbound note states the outbound subject and outbound object. A repayment plan states the repayment subject and planned period. The tool call node decides whether to invoke a plugin from the tool description and parameter descriptions; plugin parameters come from configuration, while system module parameters are already fixed.
Material updates do not trigger tool calls automatically. The model judges call timing from context semantics. Values collected by retrieval or forms do not constitute authorized execution; business verification, approval, database writes and external synchronization each require their own downstream implementation.
Hypothetical paper exercise sample
The following paper materials are used to illustrate the comparison method. They do not represent a real customer or real records.
Material one, replenishment request text: "Requesting subject: East China regional dealer A; replenishment object: model A stock; requested quantity: 300 units; request date: 10 March 2025; authorization status is pending review by brand channel manager."
Material two, outbound note text: "Outbound subject: East China regional dealer A; outbound object: model A; outbound quantity: 300 units; outbound date: 12 March 2025; outbound note status: generated, unsigned."
Draft sentence to be judged: "East China regional dealer A completed the outbound of 300 units of model A on 12 March 2025 and received repayment."
Comparison points: material one shows the replenishment request is pending review, and material two shows the outbound note is generated but unsigned. The draft sentence claims completed outbound and received repayment. No repayment record appears in the materials, and an unsigned outbound note does not constitute completed outbound. Pending review does not demonstrate a lack of operating authority; unsigned does not constitute completed outbound; an absent repayment plan does not constitute repayment received.
Constraints at the tool calls step
Tool calls need a tool description and tool parameters. The description tells the model what the tool does. Parameter descriptions and whether each parameter is required together shape whether the model calls the tool. A tool with no parameters is judged from its description alone. When a required parameter has no suitable value in context, the model may not call the tool, and prompts can guide the user to supply the value. When no parameter is required, the model may fabricate a value, so the actual request needs checking after the call.
In advanced orchestration, connect available tools to the diamond connector on the tool call module. Connected tools automatically separate tool inputs from ordinary inputs, and the description can be edited to make call timing more precise. Start with a small number of tools and tune before adding more. A tool call termination node attached after a tool ends this tool call, does not call other tools, and does not call AI to summarize the tool result. Custom tool variables extend variable inputs for nodes not treated as tool parameters; filling in the matching parameter description lets the tool call invoke that node and its downstream workflow.
Semantic recall and relevance scores do not carry permission or entity isolation. Libraries across different vector models cannot be associated together; check identifiers, exclude non-matching ones, and record the actual observation. When retrieval returns empty, a clearly defined paper input can be used to check downstream design, without claiming that a fictional rule retrieval necessarily returns empty.
Configuration and implementation checks
The table below lists configuration check items for this paper exercise. The values are hypothetical starting points for comparison against actual material values, not official defaults.
| Configuration item | Hypothetical starting point | Check method |
|---|---|---|
| Tool description | State that the comparison covers subject and status across replenishment request, outbound note and repayment plan | Check whether the description covers situations where the tool should not be called |
| Plugin parameters | Set parameter descriptions and required flags per material field | When a required parameter is missing, observe whether the model calls the tool |
| Tool connection | Connect the comparison plugin to the tool call module | Confirm the diamond connector is linked and inputs are separated |
| Tool call termination | Attach a termination node after comparison completes | Confirm no further tool calls or AI summaries occur after termination |
| Custom tool variables | Add parameter descriptions for nodes not treated as parameters | Check the parameter description matches the node actually invoked |
| Retrieval scope | Check against the target scope with access rights | Confirm the library identifier and vector model identifier match |
Three common errors
- Treating a replenishment request pending review as approved execution. The input is material one showing authorization status pending review by the brand channel manager; the comparison object is the draft sentence claiming completed outbound; the observable result is that the material contains no approval record, so the draft sentence is unsupported.
- Treating a generated but unsigned outbound note as completed outbound. The input is material two showing outbound note status generated and unsigned; the comparison object is the draft sentence claiming completed outbound; the observable result is that the signature status is absent, so the draft sentence is unsupported.
- Treating an absent repayment record as repayment received. The input is the materials containing no repayment plan or repayment record; the comparison object is the draft sentence claiming repayment received; the observable result is that repayment evidence is missing, so the draft sentence is unsupported.
How to verify the setup
- Input the text of material one and the draft sentence; compare the authorization status field with the completion status claimed in the draft sentence; the observable result is that the pending review status does not match the claimed completed outbound, so the draft sentence is unsupported.
- Input the text of material two and the draft sentence; compare the outbound note signature status with the claimed completed outbound; the observable result is that the unsigned status does not match the claimed completed outbound, so the draft sentence is unsupported.
- Input material one, material two and the draft sentence; compare whether a repayment record exists with the claimed repayment received; the observable result is that repayment evidence is missing, so the draft sentence is unsupported.
The verification above is a paper check. No tool call was actually run, and no test, deployment, approval or repayment is claimed. Fields not stated in the materials remain unknown. Pending review does not constitute approval, unsigned does not constitute completed outbound, and an absent repayment record does not constitute repayment received.