What the documents look like
This category works with post-fulfillment consumer feedback and platform service indicators. The materials come from store experience scores, marketplace service metrics, after-sales evaluations, and customer service systems. Day-to-day documents include negative review handling records, experience score breakdowns, service diagnostic reports, and rectification ledgers. These materials update daily and weekly, and they record consumer feedback after fulfillment plus platform service indicators. They do not directly determine campaign prices.
Copy drafting for product and marketing text depends on these materials as factual sources. A promotional sentence can stand only when the material states the matching subject, object, and status. A negative review handling record may show that a specific order received a specific complaint, that a service agent contacted the buyer on a later date, and that the buyer did not reply. That record does not establish a shipping-time commitment for every order in the store. The drafting step therefore needs a clear separation between what the ledger states and what a proposed sentence asserts.
Constraints at the model connections step
Copy drafting relies on a language model, and model access runs through AI Proxy. From V4.17.0, AI Proxy is a required dependency. Both the App service and the Admin service that reuses App/Service configuration must set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN. Missing either item causes startup validation to fail. Legacy OpenAI environment variables and single-model custom request addresses do not replace these two settings. AIPROXY_API_ENDPOINT takes the service root address and does not append /v1.
The system needs at least one language model and one index model to operate normally. Model capabilities are checked against the inputs the application actually uses; multimodal input, tool calling, and thinking configuration are each verified separately. Application parameters are set in the application or node configuration, and a same-named field on a compatible chat endpoint is not assumed to override orchestration settings.
Hypothetical paper test case: the negative review handling record under review reads, "On 5 March 2026, order A received a negative review with the reason 'slow shipping.' Customer service contacted the buyer on 6 March to request supporting evidence. The buyer did not reply. On 8 March, the review status was pending verification." The proposed draft sentence is, "All orders in this store ship within 24 hours." Judging that sentence requires the material to state a shipping-time limit covering all orders. The record covers only order A, its status is pending verification, and no shipping-time limit appears. The sentence is unsupported.
Configuration and implementation checks
| Configuration item | Hypothetical trial starting value | Reason and comparison method |
|---|---|---|
AIPROXY_API_ENDPOINT | Hypothetically entered as http://aiproxy:3000 | The document requires a root address with no /v1; compare the string against the documented example format and check for extra path segments |
AIPROXY_API_TOKEN | Hypothetically entered as a non-empty administrator token | The document requires a non-empty token matching the server configuration; compare character by character against the AI Proxy server record |
| Language model | Hypothetically enable one language model that passed channel testing | Copy drafting needs text dialogue; compare against the channel test result and confirm the test succeeded |
| Index model | Hypothetically enable one index model that passed channel testing | The system needs at least one index model; compare against the model enablement status |
| Model capability | Hypothetically check whether image input is supported | Check against the inputs the application actually uses; compare against the vision field in the model configuration |
| Application parameter | Hypothetically set a temperature ceiling in the application configuration | Parameters are set in the application or node configuration; compare against the value actually used in the orchestration |
Three common errors
- Entering
AIPROXY_API_ENDPOINTas a full request address that includes/v1. The input is the address string to be entered; the comparison is against the documented requirement of a root address with no/v1; the observable result is whether the string contains an extra path segment. - Substituting an application API key for
AIPROXY_API_TOKEN. The input is the token source to be entered; the comparison is against the documented requirement of an AI Proxy administrator token; the observable result is whether the token matches the server configuration. - Treating the proposed sentence "All orders in this store ship within 24 hours" as supported. The input is the negative review handling record text together with the proposed sentence; the comparison is against the subject, object, and status stated in the record; the observable result is whether the record contains a field covering all orders and a shipping-time limit.
How to verify the setup
- Take the hypothetically entered
AIPROXY_API_ENDPOINTstring and compare whether it ends as a root address without/v1. The observable result is the final form of the string. If an extra path segment appears, the value does not match the documented format. - Take the hypothetically entered
AIPROXY_API_TOKENand the AI Proxy server record, and compare whether the two are identical and non-empty. The observable result is whether the character-by-character comparison matches. If the token is empty or differs, startup validation is expected to fail. - Take the negative review handling record text and the proposed sentence, and compare whether the record states a shipping-time limit covering all orders. The observable result is whether the corresponding field appears. When the field does not appear, record the sentence as unsupported and keep the evidence gap open.