What the documents look like
Parcel handoff at the carrier scan point sits at the moment a package leaves the warehouse. The carrier scan carries responsibility, so the document set for this category is narrower than warehouse packing: electronic waybills, handoff manifests, pickup scan records and exception item registers. Sources include the carrier system, WMS shipment records, the GB/T express service standard and carrier billing. Updates arrive per batch and per day, which means a retrieval step that ignores batch or date scope can mix the status of one waybill number across different batches. This page covers knowledge base retrieval and recall for order fulfillment and after-sales answering only. A retrieval result and a value collected by a form do not constitute authorization; business verification, approval, database writes and external synchronization each require their own downstream implementation.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Knowledge bases with different vector models cannot be associated together, so the implementation checks the model identifier first and excludes bases that do not match rather than running a cross-model search to compare missed recall. Reference content is output as an array whose length may be 0, which means the downstream chain still runs even when nothing is found.
Search mode needs to be compared across semantic, full-text and hybrid retrieval for the test question, without assuming one mode suits every document. Minimum relevance takes effect only under semantic retrieval or when result reranking is used. Question optimization performs coreference resolution and question expansion, and it adds one model call before retrieval. Retrieval is recall; it does not perform permission or entity isolation.
Hypothetical paper exercise: the source text reads, "On 10 March 2025, shipment batch B-0310, carrier is site A, the handoff manifest lists 12 parcels, the pickup scan record shows 11 parcels scanned that day, and the exception item register records 1 parcel with outer packaging damage pending confirmation." The draft sentence under review is, "All 12 parcels in this batch have been picked up by site A." The subject needed for judgment is batch B-0310 and site A, the object is 12 parcels, the status is 11 scanned and 1 exception, and the valid scope is 10 March 2025. The draft sentence and the source text hold mutually exclusive values on the same subject, same period and same field, so the draft is unsupported.
Configuration and implementation checks
| Configuration item | Value or check for this page | Basis and notes |
|---|---|---|
| Associated knowledge bases | Select the bases holding electronic waybills, handoff manifests, pickup scan records and exception item registers; check that the vector model identifier matches | The node can associate knowledge bases with the same vector model; different models are not associated together |
| Search mode | Compare semantic, full-text and hybrid retrieval on the sample question | Batch and number queries can start with full-text or hybrid; natural language questions can start with semantic |
| Reference limit | For this hypothetical trial, set the starting point to the token volume needed to cover one complete handoff manifest, then compare against actual documents of differing length before adjusting | The unit is token; it controls reference content length and is not an official default |
| Minimum relevance | Label relevant and irrelevant passages in the sample first, then set the threshold | Takes effect only under semantic retrieval or result reranking |
| Question optimization | For a short question such as "what is the scan status of this batch," compare whether the subject is altered before and after optimization | Performs coreference resolution and question expansion and adds one model call |
| Rerank | When a rerank model is available and enabled by configuration, candidates are reordered a second time | When not enabled, this path does not execute |
Three common errors
- Treating a retrieval hit as authorization. Input is the sample source text and the draft sentence; compare whether the retrieval result contains an authorization field; the observable result is that the reference content holds only scan and exception records and no approval or release field, so the draft sentence remains unsupported.
- Associating knowledge bases with different vector models. Input is two bases whose vector model identifiers differ; compare whether the identifiers match; the observable result is that the mismatched base is excluded. Record the actual observation rather than asserting missed recall or redundancy.
- Treating a pending item as completed. Input is the exception register entry "outer packaging damage pending confirmation"; compare the status field value; the observable result is that the item's status is pending confirmation, and the draft sentence's "all picked up" is mutually exclusive with the value in the same field.
How to verify the setup
- Input the sample source text and the draft sentence; compare whether the subject, object, period and field of the draft match the source text; the observable result is that a mutually exclusive value in the same field is judged unsupported, and items awaiting verification remain gaps.
- Input a paper query with no matching content; compare the reference content array; the observable result is that when the length is 0 the downstream chain still runs. Do not claim that a fabricated rule search must return an empty array.
- Input a query containing an image and text; compare whether a visual model or an image-capable vector model is configured; the observable result is that the corresponding path executes according to capability, and when it is not configured that path does not execute. Do not guarantee a first hit or full recall.