What the documents look like
Short-video affiliate sales generate transactions through long-term content exposure and product link clicks. Order fulfillment and after-sales question handling often occur several days after a video is published. Customer service and operations staff need to trace back the correspondence between the linked product, the commission plan, and violation feedback from that earlier period. Data sources for this category include the selected alliance, short-video playback data, product transaction data, and platform rules. Daily documents consist of short-video scripts, linked product lists, commission plans, and violation feedback. These materials are distributed across different systems. The conversation log and audit step must restore a question-handling session or a fulfillment communication to a specific conversation, a specific dialogue entry, and a specific order context.
Constraints at the conversation logs step
Conversation identity depends on chatId for differentiation. When chatId is a non-empty string, FastGPT retrieves that conversation from the database and uses the last item in the messages array as the user question; all other messages are ignored. When chatId is not passed, the platform context is not used, and the context is constructed entirely from the passed messages. Therefore, when verifying fulfillment communication, the first step is to confirm that the chatId corresponds to the target conversation. Conversations from different entry points cannot be mixed together.
When run details are needed, call the chat completions interface with detail=true. With stream=false, intermediate values are stored in responseData. With stream=true, they are distinguished through event, where flowResponses carries the complete node responses. The log and interface return whichever fields they actually return; business data that is not returned cannot be presumed to have been fully retained.
Hypothetical paper exercise sample (paper materials, not actually executed): The short-video script original text reads "This product is a kitchen storage rack; the linked product corresponds to selected alliance product A; the commission rate follows the March plan." The linked product list original text reads "Product A, listing date March 1, status on sale." The violation feedback original text reads "The video title contains an extreme term, pending modification, not delisted." The handover request to be judged reads "Please confirm the order fulfillment status of product A based on the above materials and reply to the user that it has been handled."
Configuration and implementation checks
| Check item | Relevant setting or action for this page | Correspondence with the sample |
|---|---|---|
| Conversation identity | Pass a non-empty chatId, retrieve that conversation, and use the last message as the question | Confirm whether the question-handling conversation is the target conversation |
| Run details | Set detail=true; for non-streaming, check responseData; for streaming, check event | Inspect whether fulfillment-related node responses exist |
| Conversation list | source=api retrieves only conversations created through the API | Distinguish interface conversations from page conversations |
| Dialogue list | Retrieve paginated records by appId and chatId; loadCustomFeedbacks may be set | Review historical dialogues under that conversation |
| Single dialogue detail | Retrieve the run details of that dialogue by dataId | Locate the run record of a specific question-handling session |
| Feedback fields | Use userGoodFeedback and userBadFeedback to record feedback | Mark the quality of after-sales question handling |
This sample does not involve a numeric starting value, so no hypothetical initial value is set. Regarding authorization, authProxy for proxying team member identity must be enabled by the team owner when creating or editing the key; the proxied identity still needs permission for the target application and conversation. Retrieval and log reading do not by themselves constitute business authorization.
Three common errors
- Treating a call without chatId as having read that conversation's history. Check: The input is a request that does not pass chatId. Compare whether the returned content contains historical dialogue. If it is constructed only from the current messages, then the platform context was not used.
- Treating a detail=false return as containing intermediate values. Check: The input is a request with detail=false. Compare whether responseData appears. If it does not appear, then node-level information was not returned.
- Treating violation feedback marked as pending modification as already delisted. Check: The input is the sample violation feedback original text. Compare whether the status field reads "delisted." The original text says "pending modification, not delisted," so delisting is not supported.
How to verify the setup
- Take the sample script original text and the linked product list original text. The input is a request with chatId and detail=true. Compare whether responseData contains node records corresponding to that conversation. Only when the corresponding records appear and the chatId matches are the identity and detail items aligned.
- Take the sample handover request to be judged. The input is a request to retrieve run details by dataId. Compare whether the returned query or historyPreview contains that request content. If it contains it, the dialogue can be located; if it does not, record it as a gap.
- Take the sample violation feedback original text. The input is a paginated dialogue list request. Compare whether the corresponding feedback field appears in that conversation's records. Only when it appears and the value matches the original text can feedback retention be verified; when it does not appear, remain unknown and do not presume retention or handling.
This page addresses only: short-video affiliate sales operations / order fulfillment and after-sales question handling / conversation logs and audit. When operations staff hand over materials, actions, and requests, their authorization status must be clear; retrieval or forms cannot be treated as authorized execution.