What the documents look like
Commission settlement happens after a sale closes. The working materials are valid promotion orders, refund commission-deduction details, and settlement statements. Copy drafting for products and marketing operates on these materials, and the materials carry status and scope. A commission setting record states a rate and an effective period. A settlement statement states a billing period and an amount. A refund commission-deduction detail states the deduction target. An invoice states the issuing entity. When a draft sentence mentions commission, settlement, or deduction, it must trace back to a specific field in these materials. Rates, billing periods, and entity names that the materials do not state remain unknown and are not written as known facts.
The data updates weekly and monthly, and the documents in daily use include commission setting records, settlement statements, refund commission-deduction details, and invoices. The most time-consuming steps are checking valid orders, deducting refunded commission, and invoicing and paying. These characteristics shape what the drafting step must treat as evidence.
Constraints at the multi-turn prompts step
Multi-turn dialogue carries prior messages into the request. When chatId is a non-empty string, the system retrieves the session from the database, uses the last item of the messages array as the user question, and ignores the remaining messages. When chatId is not passed, FastGPT does not use context, and the request builds context entirely from the messages supplied. The prompt must state the material scope, the answering task, and how to handle missing evidence, so the model does not write business facts that the materials do not contain. Reply format follows the downstream consumer; selecting json_schema requires filling in the corresponding format requirement and checking the returned structure. The reply cap controls the length of a single answer, and a setting that is too low may truncate the reply. Memory turns control how many earlier rounds are referenced, and a larger value carries more content. A similarity score in retrieval results is a ranking reference and does not perform permission or entity isolation.
Hypothetical paper exercise sample. Material text: "Commission setting record: a certain product has a commission rate of 15%, effective from 1 March 2025 to 31 March 2025." "Refund commission-deduction detail: order A was refunded on 20 March 2025, deducting 15% commission." Draft sentence to be judged: "This product has a 15% commission rate, and refunded orders are also deducted at 15%." Check point: whether the subject, rate, and period in the draft sentence have corresponding fields in the source text. This sample is a paper exercise, not a record of any real customer or transaction.
Configuration and implementation checks
The table below is a hypothetical starting point for this exercise, not an official default value.
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison method |
|---|---|---|
| Memory turns | Keep a small number of history rounds | Copy drafting benefits from prior product information; compare with a pure messages request without chatId to observe the difference in carried context |
| Reply cap | Opened and set high enough to hold a complete copy sentence | A copy sentence needs to be output in full; compare with a shorter setting to observe truncation |
| Temperature | Low | Commission and deduction wording needs stability; compare with a higher setting to observe wording variation on the same input |
| Reply format | Default | Ordinary dialogue scenario; change to json_schema only when a downstream process needs to read a fixed structure, and fill in the format requirement |
| Retrieval library association | Check the library and vector model identifiers | Libraries across different vector models cannot be associated together; check identifiers first and exclude those that do not match |
| Empty-result handling | Check the downstream design against a paper input of [] | Observe how the downstream handles empty references; do not assume a fictional rule always returns [] |
Three common errors
- Treating similarity as a permission filter. A score in retrieval results is a ranking reference and does not exclude unrelated business or unauthorized records. Check: enter the same query, compare the returned list with the expected material scope, and observe whether records outside that scope appear.
- Writing an unstated rate as a known fact. When the material states only an effective period and a rate, a draft sentence that adds a billing period or an entity name has supplied a fact. Check: compare the draft sentence field by field with the source text and observe whether any field is absent from the source.
- Treating a pending review as approved. The status in a refund commission-deduction detail must be compared with the settlement statement on the same subject, the same period, and the same field; when the status is not stated, record the gap. Check: list the field values from both places and observe whether they are mutually exclusive or missing.
How to verify the setup
- Enter the material text and the draft sentence from the hypothetical sample, compare the rate and period in the draft sentence with the corresponding fields in the source, and observe whether they match. A match satisfies the condition; a field absent from the source is recorded as a gap.
- Enter a pure messages request without chatId and a request with chatId, compare the context carried in each, and observe whether history messages are ignored or used as the documentation describes.
- Enter a paper empty reference of [], compare the downstream handling design, and observe the output when the result is empty; do not claim that a fictional rule always returns empty.
This page addresses only creator commission settlement and deduction, product and marketing copy drafting, and multi-turn prompts. Whether a specific promotional sentence in product copy is supported by the materials is a separate question from enterprise identity certification.