Marketplace promotions: model connections for product questions

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

What the documents look like

Marketplace promotion registration and price management revolve around four recurring records: the campaign registration form, the promotional price approval sheet, the inventory lock sheet, and the campaign review sheet. The working data is price, stock, and campaign rhythm, updated on a weekly and per-campaign basis and reviewed jointly by campaign operations, brand sales planning, and finance business partners. The public brief marks this picture as approximate, so the exact fields follow whatever forms each organization actually uses.

For product questions and selection Q&A, model connection and configuration means bringing these materials into a searchable knowledge scope, then letting a model answer selection and material questions on that basis. Fields the materials do not state stay unknown, and the model does not fill them in.

Constraints at the model connections step

Model access depends on AI Proxy. From V4.17.0, both the App and an Admin service that reuses App/Service configuration must set AIPROXYAPIENDPOINT and a non-empty AIPROXYAPITOKEN; missing either one fails startup validation. AIPROXYAPIENDPOINT holds the service root address and does not append /v1. Legacy OpenAI environment variables or a single-model custom request address do not replace these two settings.

The system needs at least one language model and one index model to work normally. Model types cover language, index, rerank, speech synthesis, and speech recognition. The model ID is the value of the model field in the request body and is globally unique; the model name is for display only. A channel can hold a model mapping, for example mapping gpt-4o-test to gpt-4o.

Retrieval and authorization are separate. A similarity or relevance setting is not a permission filter, and a retrieval result does not mean a registration, approval, or inventory lock has been authorized. Forms collect values; business verification, approval, database writes, and external synchronization each need their own downstream implementation.

Hypothetical paper sample, for this page's checks only and pointing to no real party: the material text reads, "Campaign registration form (No. A-01) states: the registering party is a brand sales planning group, the object is a platform summer promotion, the registration stage is submitted, and the registration deadline is July 10." "Promotional price approval sheet (No. P-01) states: the approval object is the promotional price corresponding to A-01, the approval status is pending review, and no approval time is stated." The handover request to be judged is: "Please confirm that A-01 has been approved and can proceed directly to inventory lock."

Configuration and implementation checks

Configuration stays around this page's materials and primary task and lists only documented settings or implementation checks. Any value below is a hypothetical trial starting point for comparison and is not an official default.

ItemHypothetical trial starting pointCheck method
AIPROXYAPIENDPOINTFill the AI Proxy service root addressCompare with the documented requirement: no /v1 appended; a missing value fails startup validation
AIPROXYAPITOKENConfigure a non-empty admin tokenCompare with the server-side configuration for consistency; an empty value counts as unconfigured
Language modelEnable at least oneCompare with the "at least one language model" requirement
Index modelEnable at least oneCompare with the "at least one index model" requirement
Model ID and model mappingFill according to the models the materials needCompare against the models in the channel and the mapping, confirming the request model value
Material retrieval scopeInclude the four campaign material typesCompare with the target scope the operator has access to; record the gap when missing

Model status probing covers only currently enabled system models; disabled models are not probed and produce no history. The probe interval accepts an integer from 5 to 60 minutes. These settings observe model availability and do not change the authorization state of the materials themselves.

Three common errors

  1. Writing AIPROXYAPIENDPOINT as a full request address that includes /v1. The input is the configuration value, the comparison object is the documented root address, and the observable result is whether startup validation passes.
  2. Reading "pending review" as "approved." The input is the approval status field of P-01, the comparison object is the fact of approval, and the observable result is whether an approval time is stated in the material.
  3. Treating a retrieval hit as inventory lock authorization. The input is the returned material fragment, the comparison object is the authorization record in the inventory lock sheet, and the observable result is whether a lock execution record exists.

How to verify the setup

  1. For the hypothetical sample, input the original text of A-01 and P-01, compare the two fields "registration stage is submitted" and "approval status is pending review," and observe whether they share the same subject, the same period, and the same field. When no approval time is stated, the conclusion stays pending verification.
  2. For the handover request, input "Please confirm that A-01 has been approved and can proceed directly to inventory lock," compare whether the materials contain an approval record and a lock authorization record, and observe whether the request is supported by the materials. Record the evidence gap when missing.
  3. For model connection, input the configured values of AIPROXYAPIENDPOINT and AIPROXYAPITOKEN, compare whether they meet the required-field and root-address conditions, and observe whether startup validation passes. Without an actual run, describe only the comparison method and do not claim a test was performed.

A retrieval check observes returned results and failure branches and does not guarantee a first hit or full recall. An empty result can be checked against downstream design with the explicit paper input [], without claiming that a fictional-rule search necessarily returns []. Libraries across different vector models cannot be associated together; verify identifiers first and exclude those that do not match.

This page addresses only marketplace promotion registration and price management, product questions and selection Q&A, and model connection and configuration. When operations hands over work, which materials, actions, and requests sit at what authorization state cannot be treated as authorized execution on the basis of retrieval or a form.