What the documents look like
In the complaints, reports and regulatory reply category, product and marketing copy drafting draws on complaint tickets, regulatory inquiry letters, order evidence and rectification reports. These materials are produced case by case on an irregular schedule, one set per case. The people handling them are customer service supervisors, legal and compliance staff, and store owners. An error can lead to administrative investigation, platform penalties and brand reputation damage.
Compared with after-sales repair and reshipment, this category faces a platform or a regulator, so formal evidence and rectification wording are required. When drafting product or marketing copy, citation tracing requires that every promotional statement correspond to a verifiable passage in the source material. A field not stated in the material remains unknown and cannot be inferred from the practice of another category.
The core action of citation tracing is to compare each proposed sentence with the source text item by item, confirming that the subject, object, status and effective scope match. The knowledge base search node outputs citations as an array, and the length may be zero; even when nothing is retrieved, that output path still runs. An empty array does not mean the workflow terminated with an error.
The citation limit is controlled in tokens. Minimum relevance is a filtering threshold and takes effect only in semantic retrieval or when result reranking is used. Minimum relevance does not perform permission or entity isolation, and retrieval itself does not constitute authorization. Associated knowledge bases must share the same vector model before they can be associated together; this requirement is limited to the knowledge base search scope and does not extend to business registration sites or other external systems. Hybrid retrieval merges multiple result paths through RRF, and RRF is a fusion process rather than a switch that can be turned on. Image retrieval requires the corresponding image input and model capability, and reranking requires a rerank model enabled according to settings. The optional paths shown in the overall document flow diagram do not mean every request executes all of them.
Hypothetical paper exercise sample. Source passage one: "This store sold Model A thermos cups between 1 March 2025 and 31 March 2025, and the page stated a capacity of 500 millilitres." Source passage two: "The inspection report number for this batch is HC-2025-0312, and the conclusion is qualified." Source passage three: "The campaign page went live on 5 March 2025 and was taken offline on 1 April 2025." The proposed sentence to be assessed is: "Model A thermos cup, capacity 500 millilitres, passed inspection, campaign valid long term." The phrase "valid long term" is not stated in any source passage and is therefore unsupported.
Constraints at the citation tracing step
The comparison must stay on the same subject, the same period and the same field. Only mutually exclusive values in the same field constitute a contradiction. A field that is not stated, or evidence that covers only a related stage, is recorded as unsupported or pending verification, and cannot be labelled as a denial or a conflict.
The proposed sentence about the thermos cup mixes three separate source passages. Capacity comes from passage one, inspection conclusion from passage two, and campaign timing from passage three. The phrase "valid long term" has no corresponding field in any passage. Because the sources state a live date and an offline date, the statement of long-term validity is unsupported rather than contradicted. It should be recorded as pending verification, and the evidence gap should remain visible rather than being filled by inventing an authorization or effective clause.
Retrieval is not authorization. A relevance score is not a category, entity or permission filter, and no similarity setting excludes unrelated businesses or unauthorized records. Whether a retrieved citation can be used still depends on review against the effective version of the material. The requirement that associated knowledge bases share the same vector model applies only within knowledge base search. It does not mean that business registration sites or other external systems must use the same model, and it does not mean that associating libraries with different vector models produces a specific recall or redundancy outcome. The correct action is to check the vector model identifier, exclude libraries that do not match, and record what was observed.
The thinking budget is a reasoning budget and is unrelated to the citation token limit. RRF is a fusion process for hybrid results, not a configurable switch. Image retrieval and reranking are optional paths that depend on the corresponding input and model capability, and they are not executed on every request.
Configuration and implementation checks
The table below lists settings or implementation check items relevant to this feature and the primary task, using the sample above. The values are starting points for this hypothetical exercise, used for comparison and observation, and are not official defaults.
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Select the libraries holding complaint tickets and order evidence | Check the vector model identifier of each library, exclude libraries with a different vector model, and do not run cross-model retrieval |
| Search mode | Hybrid retrieval | Compare recalled content against single semantic retrieval and observe whether identifier-style wording is hit |
| Minimum relevance | Set to 0.3 | Compare with 0.5 and observe whether low-relevance fragments are filtered; effective only in semantic retrieval or reranking |
| Citation limit | Set to 2000 tokens | Compare with 1000 tokens and observe whether citation content is truncated |
| Result reranking | Enabled | Compare candidate ordering against the disabled state; a rerank model is required |
| Citation output handling | Accept the array result | Use a paper input of an empty array to check the downstream design and observe whether the path runs |
Three common errors
- Treating "valid long term" in the proposed sentence as a verified fact. Source passage three states only the campaign live date and offline date, and does not state long-term validity. The phrase is unsupported and should be recorded as pending verification, not judged as a denial or a conflict.
- Using minimum relevance to exclude unrelated businesses or unauthorized records. Minimum relevance is a filtering threshold and does not perform permission or entity isolation. Whether a retrieval result can be used still requires review against the effective version.
- Describing RRF as a switch that can be turned on, or moving the thinking budget into the role of the citation token limit. RRF is a hybrid result fusion process, and the thinking budget is a reasoning budget unrelated to the citation limit.
How to verify the setup
- Input the three source passages and the proposed sentence from the hypothetical sample, then compare each statement in the proposed sentence against the source text for the same subject, the same period and the same field. If "valid long term" has no corresponding field in the sources, record it as unsupported. The expected observation is that capacity and inspection conclusion are supported by passages one and two, while long-term validity remains unsupported.
- Input a paper empty array and observe whether the downstream answer step accepts it normally. An empty array means no content was retrieved and does not mean the workflow terminated with an error. The expected observation is that the path runs and the downstream step handles empty citations.
- Check the vector model identifier of each associated knowledge base, exclude libraries with a different vector model, and record the library list actually observed. Do not run cross-model retrieval to compare missed recalls, and do not claim that a fabricated rule retrieval necessarily returns an empty array.
This page addresses only complaints and official replies, product and marketing copy drafting, and citation tracing. Whether a specific promotional sentence is supported by the material cannot be turned into an enterprise identity certification tutorial.