What the documents look like
After-sales repair and reship work aims to restore a product to use. It does not close a case through a direct refund, which separates it from refund-only disputes. The working documents are repair requests, reship orders, inspection reports and accessory outbound orders. These materials come from the after-sales system, the order system, repair sites and the accessory warehouse WMS. Copy drafting sits inside this material set. A draft sentence about a repair cycle, a reship scope or an accessory source is checked against what those documents state. A field the material does not carry stays unknown. A similar category does not supply it.
Constraints at the model connections step
Model access depends on AI Proxy. The App service and any Admin service that reuses App/Service configuration must set AIPROXYAPIENDPOINT and a non-empty AIPROXYAPITOKEN. Missing either item causes startup validation to fail. Legacy OpenAI environment variables and a single-model custom request address do not replace these two settings. Models are created as channels and enabled on the administrator model provider page. The model ID matches the model in the channel. Language model capability, tool calling and multimodal support are checked item by item against the actual application input. Datasets that are associated together require the same embedding model; datasets on different embedding models cannot be associated together.
Hypothetical drill sample: a brand after-sales site accepts a reship request on 10 March 2026. The material text reads: "Customer order number A-20260310-017, product is a household water purifier filter element, fault description is odd taste in the output water, inspection report conclusion is that the filter element is due for replacement, reship order status is awaiting outbound from the accessory warehouse." The draft sentence to be judged is: "This brand promises to complete the reship and deliver within 48 hours." The material states the reship order status as awaiting outbound. It does not state a delivery time limit or a promising party. The draft sentence exceeds the material.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this drill | Reason and comparison method |
|---|---|---|
| AIPROXYAPIENDPOINT | Service root address filled per the actual deployment | The document requires no appended /v1; compare against the deployment record for extra path segments |
| AIPROXYAPITOKEN | Non-empty value consistent with the server side | The document requires a non-empty value; compare against the AI Proxy administrator configuration |
| Model ID | Consistent with the model in the channel | The document states the model ID is globally unique; compare against the channel model list |
| Model capability | Checked against the input the copy drafting needs | The document lists multimodal and tool calling as separately verified; compare against the sample material types |
| Probe interval | 30 minutes as a hypothetical starting point for this drill | The document supports integers from 5 to 60; compare against the alert receiver capacity, not an official default |
| Alert Webhook address | Filled per the actual receiver | The document states connectivity can be tested; compare against the receiver log |
Three common errors
- Writing a delivery time limit that the material does not carry as a brand promise. In the sample the reship order status is awaiting outbound from the accessory warehouse, and no time limit or promising party is stated. The draft sentence "complete the reship and deliver within 48 hours" has no material support.
- Replacing the AI Proxy settings with legacy OpenAI environment variables. The document states the legacy variables cannot replace AIPROXYAPIENDPOINT and AIPROXYAPITOKEN, and that missing either item fails startup validation.
- Treating a similarity setting as a permission or entity filter. The document describes retrieval as serving relevant text retrieval. It does not carry authorization or business ownership judgment, so it cannot exclude unauthorized records.
How to verify the setup
- Input the sample material text and the draft sentence "This brand promises to complete the reship and deliver within 48 hours." Compare whether the time limit and the promising party in the draft sentence appear in the material. The material states only that the reship order status is awaiting outbound from the accessory warehouse. It states no time limit and no promising party. The sentence is unsupported and is recorded as pending verification.
- Input the deployment values for AIPROXYAPIENDPOINT and AIPROXYAPITOKEN. Compare whether the token is empty and whether the endpoint has
/v1appended. When both meet the document requirements, startup validation passes. When either fails, startup validation fails and the setting is corrected per the document. - Input the product, fault description, inspection conclusion and reship order status from the sample material. Compare whether the draft sentence uses only fields the material states. If the draft sentence introduces an accessory source or repair cycle the material does not state, it exceeds the supported range. The evidence gap is retained and marked as pending verification.