What the documents look like
Wholesale stall trading runs on purchase notes, handwritten sales slips, stall lease contracts and payment receipts. These materials are updated daily or per order. On-site price negotiation and shipping arrangements mean the same batch of goods can appear with different product names, specifications, quantities and prices across different slips. Copy drafting must extract checkable product information from these materials, and the prompt must instruct the model not to add selling points that the materials do not state. Records are scattered, slips are handwritten, and fields are not standardized, so the prompt has to state the material scope and the handling of missing evidence.
For this page the working materials are purchase notes, handwritten sales slips, stall lease contracts and payment receipts, and the task is drafting product and marketing copy. The update frequency of daily or per order affects how fresh the checkable facts are, but it does not change the drafting rules themselves. The key point is that any claim placed in draft copy has to be traceable to a specific slip.
Constraints at the multi-turn prompts step
In multi-turn dialogue, memory rounds, question optimization, model context length and reply format all affect how checkable the drafted copy is. Memory rounds control how much prior conversation is carried in; a larger value makes it easier to reference earlier turns and also increases consumption. Question optimization is used for reference resolution in continuous knowledge-base dialogue, and it is not assumed to apply to every short question. The model's maximum context determines how much content a single request can reference. Reply format is chosen according to what comes next; selecting json_schema requires filling in the corresponding format requirement and checking the returned structure. For facts the materials do not state, the prompt should require the model not to add them.
Hypothetical paper exercise sample. The material text supplied for checking is three sentences: "Purchase note states: Red Fuji apple, specification 80mm, quantity 200 boxes, unit price 120 yuan per box, date 10 March 2025." "Handwritten sales slip states: Red Fuji apple, 80mm, 150 boxes, 125 yuan per box, same day." "Stall lease contract states: Stall A-12, lease term 1 January 2025 to 31 December 2025." The draft sentence to be judged is: "This stall can supply 200 boxes of Red Fuji apple today at 120 yuan per box, with a stable supply." This sentence involves two facts, quantity and price. The two slips disagree on quantity and price, and "stable supply" is not stated in the materials.
Configuration and implementation checks
| Configuration item | Related setting or check on this page | Relation to the sample |
|---|---|---|
| AI model | Select a model supporting the required context length; check maximum context and multimodal capability | The three sample sentences are short; the model must be able to read the difference between the two slips |
| Memory rounds | Keep the default or adjust for continuous follow-up; check how many history rounds are actually carried in | The sample is a single-turn check; history rounds do not change the material text |
| Reply cap | Set according to copy length; check whether the reply is truncated early | The sample draft sentence is short, so truncation risk is low |
| Temperature | Copy drafting may be set higher; check whether the output introduces content not stated in the materials | "Stable supply" is not stated in the sample, so a higher temperature requires closer checking |
| Reply format | Keep default for ordinary copy; select json_schema with the format requirement filled in when a downstream step needs a fixed structure | The sample is a natural-language draft sentence and needs no fixed structure |
| Hide AI output | Enable when a downstream node rewrites the content; check whether the downstream node receives it | The sample displays the draft sentence directly and does not need this enabled |
Three common errors
- Merging the quantity and price disagreement between the two slips into a single value. The input is the purchase note at 200 boxes and 120 yuan and the sales slip at 150 boxes and 125 yuan. Compare the same field across the two slips and observe whether the draft sentence takes only one value without noting the difference. The materials do not say which slip governs, so the difference should remain pending verification.
- Writing "stable supply" into the draft sentence although the materials do not state it. The input is the three material sentences. Compare each factual expression in the draft sentence against the material text and observe whether "stable supply" has any corresponding source. That expression is unsupported.
- Using the validity range of the stall lease contract to infer a supply period for goods. The input is the lease term stated as 1 January 2025 to 31 December 2025. Compare that period with the goods quantity and price, two different fields, and observe whether the lease term is treated as a supply commitment. The subjects and objects differ, and the two cannot be inferred from each other.
How to verify the setup
- Check whether the quantity and price in the draft sentence correspond one by one to the material text. The input is the purchase note and sales slip text together with the draft sentence. Compare the draft sentence values with the values in the two slips. If the draft sentence takes only one value and does not note the difference, the check condition is not met; if it notes the difference or keeps it pending verification, the condition is met.
- Check whether every factual expression in the draft sentence has a material source. The input is the three material sentences and the draft sentence. Compare each expression in the draft sentence with the material text. If an expression such as "stable supply" appears that the materials do not state, the condition is not met; if all expressions can be matched or are explicitly marked as not stated, the condition is met.
- Check whether different fields are inferred from each other. The input is the lease term in the stall lease contract and the goods quantity and price. Compare the lease term field with the supply fields. If the draft sentence uses the lease term as a basis for a supply period, the condition is not met; if the two kinds of fields are stated separately without inference, the condition is met. All of the above are paper checks; no actual dialogue or system operation was executed.