What the documents look like
Central chain procurement and distribution centers on headquarters purchasing and outbound delivery, not on store-level display checks. The working materials are centralized procurement contracts, replenishment plans, delivery notes and store requisitions. Operational data arrives from ERP purchasing, WMS, store POS and supplier delivery records, with daily and weekly update cycles.
Document comparison in this setting asks one narrow question: for the same contracting entity, the same period and the same field, do the contract terms and the transaction records agree? Tool calls and plugins organize that question into steps the model can select. The model reads a tool introduction and parameter introductions, then decides which comparison action to invoke. The tool does not certify the business relationship; it performs the comparison the workflow exposes to it.
A comparison is only meaningful when subject, period and field line up. Where the same field carries two mutually exclusive values, that is a contradiction. Where a value is not stated, or where only evidence of a related stage exists, the conclusion stays unsupported or pending verification. A pending review does not show that a party lacks operating authority. Signing, reaching cooperation and contract effectiveness are three separate facts. A quotation validity period does not state a supply period.
Constraints at the tool calls step
A tool consists of a tool introduction and tool parameters. System module parameters are already fixed; plugin parameters are configurable. Two parameter values matter most: the parameter introduction and whether the parameter is required. When a suitable required parameter is missing, the model may decline to call the tool, and a prompt can guide the user to supply it. When no parameter is required, the model sometimes fabricates one.
In advanced orchestration, available tools are connected to the tool call module, and the introduction can be edited so the call timing is more precise. The documented guidance is to start with a small number of tools and tune before adding more. The tool call termination node ends the current call: no further tools are called and the AI does not summarize the tool result. Custom tool variables extend the variable inputs for nodes that are not treated as tool parameters, so the tool call can reach those nodes and the workflow after them.
Hypothetical paper exercise sample. The centralized procurement contract text supplied for checking reads: "Party A: a chain supply chain center; Party B: a food supplier. Contract number HZ-2024-001, valid from 1 January 2024 to 31 December 2024, agreed monthly supply volume 500 cases." The delivery note text reads: "Delivery note number PS-2024-0301, receiving store: a store, delivery date 5 March 2024, delivery quantity 480 cases, corresponding contract number HZ-2024-001." The draft sentence to be judged reads: "In March 2024 the delivery quantity under this contract matched the agreed monthly supply volume." The contract states 500 cases and the delivery note states 480 cases. The same field carries different values, so the draft sentence is unsupported. This is a paper exercise; no run has been performed.
Configuration and implementation checks
| Configuration item | Content for this page | How to check |
|---|---|---|
| Tool introduction | State the purpose of comparing the centralized procurement contract with the delivery note, the input conditions, and the situations where the tool should not be called | Check that the introduction covers contract number, period and quantity field |
| Tool parameters | Contract number, period and quantity are configurable plugin parameters; review each parameter introduction | Compare each parameter introduction with its required flag |
| Required parameter | Contract number is set as a required parameter | When it is missing the model may not call the tool; guide completion and check the actual request |
| Tool connection | Connect the contract comparison tool to the tool call module | Check that the connection point and the edited introduction take effect |
| Tool call termination | Place it after the comparison tool to end the current call | Check that no further tool and no AI summary follow |
| Custom tool variable | Add a variable description for a node not treated as a tool parameter | Compare the variable description with the actual input |
Hypothetical starting point for this exercise: contract number is set as a required parameter, because the sample links contract and delivery note through the contract number. This is compared against the observed call behavior when no required parameter is set. The starting value is chosen for this paper exercise only and is not an official default.
Three common errors
- Treating a parameter introduction as a guarantee of execution. The parameter introduction describes what the parameter means. Whether the model calls the tool, and whether the call succeeds, must be checked against the actual request and response.
- Treating a retrieval hit as an authorization state. A retrieved contract or delivery note shows only that the material was recalled. It does not show that the comparison ran or that a handover was approved.
- Treating a pending review as missing operating authority. In the sample the draft sentence is pending judgment, which shows only that the conclusion is unsupported. It does not establish that any party lacks authority.
How to verify the setup
- Input the sample contract text and delivery note text. Compare the agreed 500 cases with the 480 cases on the delivery note. If the tool returns that the two quantities differ, the comparison action ran as intended. If it returns that they agree, check whether a parameter was fabricated.
- Input the draft sentence "In March 2024 the delivery quantity under this contract matched the agreed monthly supply volume." Compare the draft sentence with the sample source text. If the tool output is unsupported or pending verification, the condition is met. If the output states agreement, the condition is not met.
- Input a request that omits the contract number. Compare whether the model calls the comparison tool. If the model does not call it and asks for the contract number, the required parameter setting is working. If the model fabricates a contract number, check the parameter introduction and the required flag.
All three checks are paper expectations. No run, deployment or approval has taken place. Where evidence is absent, it remains absent.