What the documents look like
Marketplace advertising adjusts bids against traffic and conversion in near real time, and it does not centre on platform campaign review. The comparison materials for this category are campaign plans, keyword reports, audience package descriptions and conversion attribution tables. These materials come from sources such as Juliang Qianchuan, JD Express, Alimama and store transaction data, and their update frequency is daily or hourly.
For the contract and transaction material comparison task, tool calls and plugins need to read the subject, object, status and validity scope recorded in those documents so that operations staff can check the authorization state at handover. A retrieved value or a value collected by a form does not constitute authorization execution. Business verification, approval, database writes and external synchronization each require their own downstream implementation.
Two features of this category shape the task. First, campaign plans and keyword reports move on a daily or hourly cycle, so a comparison run can encounter a plan status and a report period that were recorded at different moments. Second, the materials are not uniform in structure: a campaign plan carries a confirmation status, a keyword report carries a coverage period, and a conversion attribution table carries its own field set. Tool descriptions and parameter descriptions have to name the fields being read, because the model decides whether to call a tool from the description, the parameter descriptions and whether a parameter is required.
Constraints at the tool calls step
A tool description tells the model what the tool does, so it should state the input conditions and the situations in which the tool should not be called. System module parameters are already fixed, while plugin parameters are configurable; the two are checked separately, and a parameter description does not amount to a guarantee of execution. When a suitable required parameter is absent from the context, the model may not call the tool, so the interaction has to supply the parameter and the actual request is then checked.
In advanced orchestration, available tools are connected to the tool call module, and a small number of tools is used for tuning before more are added. The tool call termination node ends the current tool call; after it runs, no further tools are called and the AI is not asked to summarize the tool result. A custom tool variable extends the variable input for a node that is not treated as a tool parameter or cannot be called as a tool, by filling in the corresponding parameter description.
Retrieval carries its own limits. A knowledge base that shares the same vector model is associated only under the documented condition, and that condition does not extend to business registration sites or other external systems. Semantic recall and relevance scores do not perform permission or entity isolation. Retrieval is not authorization.
Hypothetical paper exercise sample. The material text reads: "Party A: a marketplace store; Party B: an outsourced operations service provider; contract period: 1 March 2025 to 31 August 2025; campaign plan confirmation status is pending review; keyword report coverage period: 1 March 2025 to 31 March 2025." The draft sentence to be assessed reads: "The campaign contract between the store and the outsourced operations service provider took effect on 1 March 2025, and the keyword report coverage period matches the contract period." In this sample the contract effective date is not stated, the pending review status does not show an absence of operating permission, and the keyword report coverage period does not match the contract period.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this page | Reason and comparison method |
|---|---|---|
| Tool description text | State that the tool reads the subject, object and status in contract and transaction materials | The description has to cover input conditions and situations where the tool should not be called, and is compared field by field against the material |
| Tool parameters | Contract period, campaign plan status, keyword report coverage period | System module parameters and plugin parameters are checked separately; when a required parameter is missing the model may not call the tool |
| Tool connections | Connect only the contract material reading tool and the keyword report reading tool | A small number of tools is tuned first, rather than adding many at once |
| Tool call termination | Place the termination node after contract material reading | After termination no further tools are called and the AI is not asked to summarize the result |
| Custom tool variable | Parameter descriptions filled in as contract status and report coverage period | Extends input for a node not treated as a tool parameter; the actual request is checked |
| Retrieval scope | Limited to a target retrieval scope with access rights | When it is missing, the gap is recorded; cross-model retrieval is not run to compare missed recall |
Three common errors
- Treating a pending review status as an effective contract. The input is the sample line "campaign plan confirmation status is pending review"; the comparison is whether the material states a contract effective date; the observable result is that the material states no effective date, so the draft claim is not supported.
- Treating the keyword report coverage period and the contract period as one field. The input is the two separate periods in the sample; the comparison is the field name and the value; the observable result is that the coverage period runs from 1 March to 31 March 2025 while the contract period runs from 1 March to 31 August 2025, so the two are not consistent.
- Treating a retrieval result as authorization execution. The input is a retrieved contract material fragment; the comparison is whether it contains an approval or signature record; the observable result is that the retrieval result contains text only, so retrieval does not constitute authorization.
How to verify the setup
- Input the sample line "campaign plan confirmation status is pending review" together with the draft sentence "the contract took effect on 1 March 2025". Compare whether the material states an effective date. If no effective date is stated, the not-supported finding is satisfied and the effective finding is not.
- Input the keyword report coverage period and the contract period from the sample. Compare the start and end dates of the two fields. If the dates differ, the not-consistent finding is satisfied; if they are the same, the field names are checked further to see whether they are the same field.
- Input a paper empty result
[]. Compare whether the downstream design carries the empty result as a pending-verification state. If the downstream design marks it as pending verification, the condition is satisfied; if it marks it as a negative or conflicting finding, the condition is not satisfied.