What the documents look like
Advertising on marketplace traffic adjusts bids against traffic and conversion in near real time, and campaign approval by a platform promotion review is not the governing gate. The working inputs for copy drafting are therefore the delivery plan, keyword reports, audience package descriptions and conversion attribution tables. These materials update at day and hour frequency, so a bid, audience definition or attribution basis behind a single sentence may already have changed while the draft is still open. Whether a promotional sentence is supported by the material depends on whether the material states the same subject, the same period and the same field with a matching value. It does not depend on how much delivery data is available.
The form input node and the user selection node are user interaction nodes. When either node is triggered, the conversation enters an interaction state, the workflow state is recorded, and execution continues after the user completes the interaction. The form input node collects the information the user fills in and passes it to later nodes; it does not carry a business database function of its own. The user selection node evaluates the user's choice and executes the corresponding branch. Field validation, business approval, write-back and batch operations are implemented in an external system or in later nodes as needed.
Constraints at the interactive forms step
The interaction nodes solve collection and branching. They do not decide whether material supports a promotional sentence. The object of comparison in copy drafting is the relationship between a proposed sentence and the source wording, and the interaction nodes only gather the information to be compared and hand it to later nodes.
Hypothetical drill sample. Source wording one: "This store ran a spend-and-save promotion on model A goods from 1 March 2026 to 31 March 2026, spend 300 yuan and save 40 yuan, and the promotion target was store members." Source wording two: "In the same period, the daily budget for model A goods in the feed delivery plan was 500 yuan, and the plan status was paused." Proposed sentence to be judged: "Model A goods is currently running a spend 300 yuan and save 40 yuan promotion, and all users may take part."
In this sample the promotion target is store members, while the proposed sentence says all users may take part. These are mutually exclusive values in the same field, so they form a contradiction. The plan status is paused, which belongs to a different field from whether the promotion is running, so it cannot be used to conclude that the promotion does not exist. The material does not state whether the promotion has ended, and it does not state whether non-members may take part. Both points remain unsupported.
Configuration and implementation checks
| Configuration item | Hypothetical trial starting value | What it is compared against |
|---|---|---|
| Fields collected by the form input node | Proposed sentence text, source wording, material period | The interaction state does not reach completion when any field is missing |
| Branch condition of the user selection node | The "yes" branch when proposed sentence and source wording agree in the same field | The "no" branch when the same field holds mutually exclusive values, and the conflict is recorded |
| Required information check | Subject, target, status, effective scope | A missing item is listed as pending verification, not as a denial |
| Division of downstream work | Field validation and write-back are implemented by an external system | The interaction nodes themselves do not perform approval or database writes |
| Empty-result handling | Paper input [] is handed to downstream design for checking | No claim that a fictional rule retrieval returns [] |
These starting values are the values for this hypothetical trial and are not official defaults. The reason for choosing them is that judgement in this sample depends only on subject, target, status and effective scope; adding fields would introduce checks unrelated to the primary task.
Three common errors
- Writing "the material does not state whether non-members may take part" as "non-members may not take part". A point that is not stated is unsupported, and it is not a denial.
- Treating "the plan status is paused" as evidence that the promotion has ended. Plan status and promotion status are different fields, and one cannot stand in for the other.
- Treating the branch result of the user selection node as a business approval result. The node executes a branch, while approval is implemented in an external system or in later nodes.
How to verify the setup
- Enter the proposed sentence and source wording one from the hypothetical sample. Compare the promotion target field: the source says store members, and the proposed sentence says all users. The observable result should be that the "no" branch is taken and the contradiction is recorded, without reaching the completion state.
- Enter source wording two and the proposed sentence from the hypothetical sample. Compare the object field: source wording two describes plan status, and the proposed sentence describes the scope of promotion participation. The observable result should be that the two do not form a same-field comparison, and whether the promotion has ended is listed as pending verification.
- Enter paper empty result [] as the retrieval hand-off. Compare whether the downstream design has a defined treatment for empty input. The observable result should be that the gap is recorded, without claiming that a fictional rule retrieval returns empty or returns a hit.