What the documents look like
Pinduoduo low-price campaign operations revolve around campaign resource slots, campaign inventory, and refund-rate fluctuation. The working materials named in the brief are the campaign registration form, the product cost calculation sheet, the after-sales refund slip, and the platform in-site message. These come from the merchant backend, campaign resource-slot data, ERP inventory, and logistics signed-receipt data, and they update daily or per campaign.
For contract and transaction material comparison, the practical work is placing the price and inventory commitment in the registration form, the margin basis in the cost sheet, and the actual settled amount in the refund slip side by side under the same operating entity, the same campaign period, and the same field. The platform in-site message may carry a campaign rule change or a registration result; it is a notification material awaiting verification and does not by itself act as authorization or proof that a contract took effect.
Two features of this category shape the comparison task. Campaign price and inventory caps are numeric fields that must match across documents, so citation tracing has to show which original sentence supports each number. Refund-rate fluctuation means the settled amount in a refund slip can differ from the campaign price in the registration form, and the comparison must keep those two stages separate rather than treating one as evidence for the other.
Constraints at the citation tracing step
The knowledge base search node can associate one or more knowledge bases that use the same vector model, and its output is a citation array whose length may be zero. A zero-length array means no content was retrieved on that path; it does not mean the workflow stopped, and it does not mean any approval occurred. The citation content is then handed to the answering step, where the answer must be checked line by line against the cited original text.
The similarity score and the minimum relevance setting act only on content relevance. They do not perform permission filtering or entity isolation, so a relevance threshold cannot be relied on to exclude unrelated businesses or unauthorized records. Retrieval is not authorization. The same-vector-model requirement applies to the scope of knowledge base association and does not extend to business-registration websites or other external systems. Hybrid retrieval merges multiple result paths through RRF, reranking requires a model and must be enabled by configuration, and image retrieval requires image input plus model capability; these paths are not executed on every request.
Hypothetical paper exercise sample. Material original text one: "Party A commissions Party B to supply product A during campaign period XX, at a campaign price of 9.9 yuan per unit, with an inventory cap of 500 units, and the quotation is valid until the day before the campaign starts." Material original text two: "Party B confirms that the cost of product A is 8.5 yuan per unit, including 1.2 yuan in logistics cost, and settlement during the campaign is based on the actual signed-receipt quantity." The draft sentence to be judged: "Party B has agreed to supply 500 units of product A at 9.9 yuan per unit, the contract has taken effect, and Party A may place the order immediately." This draft sentence merges the quotation validity period, the inventory cap, and contract effectiveness into established facts, and it lacks signing and effectiveness evidence. This is a paper exercise only; no test, approval, or business action has occurred.
Configuration and implementation checks
| Configuration or check item | Hypothetical starting point for this exercise | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Select the same-vector-model knowledge bases holding the campaign registration form and the cost calculation sheet | Compare the vector model identifier of each selected base, exclude bases with a different model, and record the actual association scope |
| Search mode | Use hybrid retrieval | The materials contain numeric fields such as campaign price and inventory cap; compare whether the citation array, after RRF merging with semantic recall results, covers the original text |
| Citation cap | Set a hypothetical starting point equal to the token volume needed to cover both full material texts | Compare against the actual length of the originals and confirm the citation array includes campaign price, inventory cap, and settlement basis |
| Minimum relevance | Set a hypothetical starting point at a middle value | Compare against the citation results, observe whether low-relevance fragments are filtered, and keep citations that map line by line to the originals |
| Citation output handoff | Pass the citation array to the answering step | Check whether each answer sentence corresponds to the original text, and whether the contract-effectiveness wording in the draft has support in the citations |
| Material version check | Use the current and historical versions of the campaign registration form | Compare the campaign price and inventory cap fields across versions and confirm the cited source version matches the current campaign period |
Three common errors
- Treating retrieval output as authorized execution. The knowledge base search node outputs a citation array whose length may be zero; an empty array only means no content was retrieved and does not mean approval or contract effectiveness. Check by feeding the citation array from the hypothetical sample, comparing whether it contains signing or effectiveness fields, and observing whether "the contract has taken effect" in the draft has corresponding original text in the citations.
- Treating the similarity score as a permission or entity filter. The minimum relevance setting only filters low-relevance content and does not exclude unrelated businesses or unauthorized records. Check by feeding the campaign registration form for the same campaign period together with a refund slip from a different campaign, comparing whether the citation results mix materials from different entities, and observing whether each citation corresponds to the same campaign period.
- Treating the quotation validity period as a supply term or contract effectiveness. In the hypothetical sample the quotation is valid until the day before the campaign starts and the inventory cap is 500 units; these fields do not show that Party B agreed to supply 500 units or that the contract took effect. Check by feeding the draft sentence and the material originals, comparing whether "has agreed," "has taken effect," and "may place the order immediately" have support in the originals, and observing whether missing fields are recorded as pending verification.
How to verify the setup
- Feed material original text one, material original text two, and the draft sentence from the hypothetical exercise. Compare whether the citation array contains the campaign price, inventory cap, quotation validity period, and settlement basis, and observe whether the answer sentences correspond line by line to the originals. When the contract-effectiveness and immediate-order wording in the draft has no corresponding content in the citations, record it as unsupported.
- Feed an explicit paper input of an empty citation array []. Compare whether the downstream answering step handles it as an empty result and observe whether the workflow continues to execute. Do not claim that a fictional rule search necessarily returns an empty array.
- Feed two knowledge base association requests with different vector model identifiers. Compare whether the association scope is limited to same-model bases, observe whether different-model bases are excluded, and record the actual association result. Do not run cross-model retrieval to compare missed recall, and do not infer the runtime result of a joint association.