Marketplace promotions: model connections for copy drafting

Review model connections for copy drafting in Marketplace promotions. Check document scope, failure paths and source evidence through controlled tests.

What the documents look like

Platform promotion enrollment and price management runs on four recurring documents: the campaign enrollment form, the promotional price approval sheet, the inventory lock sheet, and the campaign review sheet. The fields that matter for copy drafting are price basis, inventory quantity, and campaign timing. A draft sentence that cites a price, a quantity, or a date range must trace back to one of these documents, and any field the documents do not state stays unknown.

The public brief describes this category as centered on price, inventory, and campaign rhythm rather than advertising-language review. It lists weekly and per-campaign update frequencies. This page does not turn those frequencies into batch synchronization, full synchronization, real-time triggering, or transaction-triggered behavior. The brief is marked as roughly accurate, so the document set is treated as a working description rather than a verified inventory.

Constraints at the model connections step

Model access depends on AI Proxy. Both the App service and any Admin service that reuses App/Service configuration must set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN; missing either one causes startup validation to fail. The endpoint holds the AI Proxy service root address and does not carry /v1. The system needs at least one language model and one index model to operate.

A model ID is the value of the model field in the request body and is globally unique. Model mapping takes the model FastGPT requests and maps it to the model actually sent upstream; the documented example maps gpt-4o-test to gpt-4o. Model status probing covers only currently enabled system models. After a first probe failure the system retries at most three times, so a single round makes at most four attempts; only when all four attempts fail is the model judged abnormal. Retrieval and relevance settings do not carry permission or entity-isolation duties, and retrieval is not authorization.

For this page the drafting task is checked against source wording. A similarity score, a retrieval association, or a model connection does not verify that a promotional price was approved, that inventory was locked, or that a campaign period is valid.

Hypothetical paper test case. The material text is: "The enrollment period for this campaign runs from March 1 to March 7. The promotional price approval sheet states a final price of 199 yuan, applicable to product A, the insulated cup. The inventory lock sheet states a campaign inventory of 500 units for that product." The draft sentence to be judged is: "Product A insulated cup, final price 199 yuan during the campaign, inventory 500 units, original price restored after the campaign ends." The post-campaign price state is not stated in the material.

Configuration and implementation checks

Configuration itemHypothetical trial starting valueReason and comparison method
AIPROXYAPIENDPOINTThe AI Proxy service root addressDocumentation requires the root address without /v1; compare with the documented example format
AIPROXYAPITOKENA non-empty token matching the server-side valueDocumentation requires non-empty and matching; distinguish it from an application API key
Language modelOne enabled language model on a tested channelDocumentation requires at least one; compare with the channel test result
Index modelOne enabled index model on a tested channelDocumentation requires at least one; compare with the channel test result
Model mappingA mapping object entered per channelDocumented example maps gpt-4o-test to gpt-4o; compare with the upstream actual model name
Model status probingHypothetical trial starting value set to enabled, interval entered as an integer from 5 to 60 minutesDocumentation gives that range; compare with status records on the probe panel

These values are a hypothetical trial starting point for this page, not documented defaults. The documented defaults for the two AI Proxy variables are absent, and the documentation marks them as required.

Three common errors

  1. Entering a full request address that includes /v1 as AIPROXY_API_ENDPOINT. Input: the configuration value. Comparison: the documented requirement for a service root address without /v1. Observable result: startup validation or a model call fails, and the failure is recorded as observed rather than assigned a cause beyond the documentation.
  2. Using an application API key in place of AIPROXY_API_TOKEN. Input: the token field. Comparison: the documented requirement for a non-empty administrator token matching the server-side value. Observable result: authentication does not pass, and the actual response is recorded.
  3. Writing "original price restored after the campaign ends" into the draft sentence. Input: that sentence. Comparison: the hypothetical material text, which states the enrollment period, the 199 yuan final price, product A, and 500 units, and says nothing about the post-campaign price state. Observable result: that phrase is unsupported by the material and remains unsupported.

How to verify the setup

  1. Take the hypothetical material text and the draft sentence. Compare "199 yuan," "500 units," and "March 1 to March 7" against the material wording. When the price and quantity expressions match the material, the price and quantity statements meet the condition. The phrase about restoring the original price remains unsupported, and the missing post-campaign price state stays missing.
  2. Take the channel configuration and the model mapping. Compare the model identifier before and after mapping. When the mapping object corresponds to the upstream actual model name, the condition is met. When it does not correspond, record the actual observation. A mapping entry does not by itself confirm that the upstream model answered.
  3. Take the probe panel records. Compare the status and the event type. When model_status_error or model_status_recovered appears, interpret it according to the documentation. When neither appears, record that nothing was observed and do not assert a reason. A probe result concerns model availability and does not verify promotional price approval, inventory locking, or campaign authorization.

Across all three checks, an unsigned termination is not a completed termination, a pending approval request is not approval, and an offered option is not execution. Where the documents do not state a field, the field remains unknown.