What the documents look like
Shopify standalone-store order fulfillment differs from TikTok Shop US in one decisive way: the merchant configures payment, shipping and fulfillment services, and no platform-wide automatic return process applies. The operational records for this task are order details, shipping-rate templates, shipping labels and refund records. They originate from Shopify Admin, payment gateways, carrier systems and inventory-location data, and they update per order and daily.
For cross-border materials and multilingual retrieval, each of those records may exist in more than one language version. Citation tracing requires that every quoted item can be returned to its original document, original field and original version. A retrieval result supplies candidate content only. It does not confirm order status, refund eligibility or fulfillment authorization. Authorization decisions, refund execution and fulfillment-status synchronization require downstream system implementation.
Constraints at the citation tracing step
The knowledge-base search node can select one or more knowledge bases that share the same vector model, and it outputs cited content as an array whose length may be zero. The citation cap is controlled in tokens. The minimum relevance value takes effect only under semantic retrieval or result reranking. Hybrid retrieval merges multiple result paths through RRF. Reranking requires a ReRank model and applies according to its configuration. Image retrieval requires image input and model capability. These mechanisms confirm that search output is candidate citation, while authorization judgment, refund execution and fulfillment-status synchronization belong to downstream systems.
Hypothetical paper exercise sample: a standalone-store merchant holds a refund record whose source text reads: "Refund RF-2026-0412, linked order SO-8841, refund amount USD 120, refund status is submitted to payment gateway, gateway receipt status: processing. Refund request date 12 April 2026." A separate handover draft sentence reads: "Refund RF-2026-0412 is complete; order SO-8841 can be closed." The question to decide: does the source text support the phrase "refund is complete" in the draft sentence?
Configuration and implementation checks
| Configuration item | Hypothetical trial starting point | Reason and comparison method |
|---|---|---|
| Linked knowledge bases | Select the Chinese refund-record base and the English refund-record base that use the same vector model | Only identical vector models can be linked together; verify the vector model identifier first and exclude any base that does not match |
| Search mode | Hybrid retrieval | Refund number RF-2026-0412 is an identifier-type exact query; hybrid retrieval uses semantic recall and full-text recall together |
| Citation cap | Set to 2000 tokens as the hypothetical trial starting point | This value must be compared against the actual chunk lengths of the refund-record documents; the documents state no default value |
| Minimum relevance | Set to 0.5 as the hypothetical trial starting point | This value takes effect only under semantic retrieval or result reranking; compare it against the rerank score distribution before adjusting |
| Result reranking | Enabled for this hypothetical trial | Reranking requires a ReRank model; once enabled it filters by a 0-1 score, to be compared against ordering without reranking |
| Question optimization | Not enabled for this hypothetical trial | The refund-number query is complete; question optimization adds a model call and latency, to be compared against retrieval results with it enabled |
Three common errors
- Treating retrieval output as an authorization conclusion. Observable error: after retrieving refund record RF-2026-0412, the draft sentence states "refund is complete; order SO-8841 can be closed." The source text states only "refund status is submitted to payment gateway, gateway receipt status: processing" and records no completed gateway refund. Check: input the refund-record source text and the draft sentence, compare the status word in the draft sentence against the status fields in the source text, and observe whether the draft sentence adds a completion state the source text does not record.
- Linking knowledge bases that use different vector models. Observable error: the Chinese refund-record base and the English refund-record base are added to the same search node. Check: input the vector model identifiers of the two knowledge bases, compare whether the identifiers match, and observe whether the search node permits joint linking when they do not match.
- Treating an empty citation array as a workflow failure. Observable error: when the search node outputs an empty array, the flow is judged failed and terminated. Check: input the paper value [] as the search node output, compare whether the downstream node continues under its empty-citation design, and observe whether the downstream node handles the empty array as a continuable state.
How to verify the setup
- Input the refund-record source text from the hypothetical exercise sample and the draft sentence "Refund RF-2026-0412 is complete; order SO-8841 can be closed." Compare "refund is complete" in the draft sentence against "refund status is submitted to payment gateway, gateway receipt status: processing" in the source text. Observe whether the draft sentence is judged unsupported by the source text. Expected: the completion state in the draft sentence has no corresponding field value in the source text, so the judgment is pending verification.
- Input the vector model identifiers of the Chinese refund-record base and the English refund-record base. Compare whether the two identifiers are identical. Observe whether the search node permits joint linking. Expected: when the identifiers differ, joint linking is not permitted; record the actual observation.
- Input the paper value [] as the search node output. Compare whether the downstream node continues under its empty-citation design. Observe whether the downstream node treats the empty array as a continuable state. Expected: the empty array does not trigger workflow termination, and the downstream node continues as designed.