What the documents look like
Brand-owned stores are staffed and accounted for directly by the brand; they are not managed through franchise contracts. Product information and selection questions draw on POS systems, member CRM, ERP inventory, and store inspections. The everyday documents are daily sales reports, shift schedules, display photos, and inventory transfer orders. These materials update daily or weekly. An answer about product selection must trace back to a specific store, a specific product, and a specific date range. Inventory figures and sellable status in an answer must match the basis used in the transfer order or sales report being cited. Fields the materials do not state stay unknown; practice in other stores does not fill them in.
Constraints at the model connections step
From V4.17.0, AI Proxy is a required dependency. The App and any Admin service reusing App/Service configuration must set AIPROXY_API_ENDPOINT and a non-empty AIPROXY_API_TOKEN; missing either one causes startup validation to fail. Older OpenAI environment variables or a single-model custom request address cannot replace these two settings. AIPROXY_API_ENDPOINT holds the AI Proxy service root address and does not append /v1. The system needs at least one language model and one index model to work. A model ID is the value of the model field in the request body and is globally unique; a model name is for display and can be customized. Model mapping connects the model FastGPT requests to the actual upstream model, for example gpt-4o-test mapped to gpt-4o. Model call logs are retained for 1 hour and this is configurable through environment variables. Retrieval finds relevant text chunks; it does not carry permission or entity isolation. Libraries across different vector models cannot be associated together; identifiers must be checked first and non-matching items excluded.
For this page's primary task, the following hypothetical drill sample applies. A regional manager for a brand-owned store submits a selection handover request. Attached are a daily sales report excerpt, "March 1 to March 7, 2026, Store A women's dresses: size S sold 12 units, size M sold 3 units," and an inventory transfer order excerpt, "March 8, 2026, Store A transfers out 5 units of dress size M to Store B, status is pending review." The proposed judgment sentence is, "Store A has no dress size M in stock; recommend transferring back from Store B." Checking compares only the same store, the same product, and the same field within the same period.
Configuration and implementation checks
The items below concern the product information and selection question task. The starting values are chosen for this hypothetical drill and are used to compare against what the page displays; they are not official defaults.
| Configuration item | Hypothetical drill starting value | How to check |
|---|---|---|
AIPROXY_API_ENDPOINT | The AI Proxy service root address, without /v1 | Compare character by character with the root address form in the environment variable documentation |
AIPROXY_API_TOKEN | A non-empty administrator token | Confirm it matches the server configuration; an application API key is not a substitute |
| Language model | At least one enabled | View enablement status on the model provider page |
| Index model | At least one enabled | View enablement status on the model provider page |
| Model mapping | Filled per the model ID used in the sample | Compare against the mapping relation in the channel configuration |
| Channel test | Run model test on the configured channel | View each model's test result and request duration |
Three common errors
- Treating "pending review" on a transfer order as a completed transfer. The input is the transfer order excerpt in the sample; the comparison object is the status field value "pending review"; the observable result is that the proposed judgment sentence declares size M out of stock. Pending review does not show that the transfer was executed.
- Using sales figures from a different period to infer current inventory. The input is the units sold from March 1 to 7 in the sample; the comparison object is "out of stock" in the proposed judgment sentence; the observable result is that the two cover different periods and different fields. Units sold are not an inventory balance.
- Treating retrieval results as a basis for authorization. The input is the text chunk returned by the selection question; the comparison object is the store and product scope the handover request requires; the observable result is that the chunk does not state that store's transfer authority. Retrieval does not perform permission filtering.
How to verify the setup
- Take the daily sales report excerpt and the proposed judgment sentence as input. Compare whether "March 1 to March 7, 2026" and "out of stock" are the same period and the same field. The observable result is that both the period and the field differ, so the judgment is unsupported.
- Take the transfer order excerpt and the proposed judgment sentence as input. Compare whether the status field "pending review" and "out of stock" are mutually exclusive. The observable result is that the status does not state completion, so the judgment is pending verification and the evidence gap is kept.
- Take the value entered for
AIPROXY_API_ENDPOINTand the root address form in the environment variable documentation as input. Compare whether/v1was appended. The observable result is that an appended/v1does not match the documented form. When the entered value is empty, startup validation fails, which is an unknown configuration state.