What the documents look like
Peak-sale warehouse and delivery scheduling centers on short-term peak capacity and the arrangement of people, vehicles, and space. The rhythm is per campaign and daily. Historical peak-sale orders, platform campaign registrations, WMS capacity, and courier pickup capability form the main data sources. Day-to-day documents include peak-sale forecast tables, staff rosters, courier assurance plans, and exception daily reports. These materials commonly mix Chinese and foreign-language warehouse and delivery terminology, identifiers, and proper nouns, so retrieval has to cover natural-language questions and exact identifiers at the same time. The cross-border material and multilingual retrieval task requires that once these documents enter a knowledge base, the relevant passage can be recalled for a question and passed to later human review.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Search mode can be semantic retrieval, full-text retrieval, or hybrid retrieval; hybrid retrieval merges the two result sets through the RRF formula, and RRF is a fusion process, not a switch. The citation limit controls, in tokens, the maximum content cited per search. The minimum relevance score is a 0 to 1 filter threshold that takes effect only under semantic retrieval or when result reranking is used, and it does not carry permission or entity-isolation duties. Question optimization performs coreference resolution and question expansion before retrieval and makes an additional model call. Image retrieval requires the corresponding image input and model capability, and reranking requires a model and must be enabled according to settings; these paths are not mandatory steps of every request. Knowledge bases built on different vector models cannot be associated together, so identifiers should be checked first and non-matching bases excluded.
Hypothetical paper exercise sample: the source material for review reads, "Peak-sale forecast table (hypothetical identifier A-01) states: from October 20 to October 25, the temporary staff scheduling cap at the Hangzhou warehouse is 120 people per day; the table was submitted by the warehouse and delivery planning role on October 15, and its status is pending review." A draft sentence to be judged reads, "A-01 has confirmed that the Hangzhou warehouse can arrange 120 people per day from October 20." The subject needed for the judgment is A-01 and the Hangzhou warehouse, the object is the temporary staff scheduling cap, the period is October 20 to October 25, and the status is pending review. The draft sentence writes the pending-review status as confirmed, which is an unsupported statement.
Configuration and implementation checks
The configuration below is built around that sample and lists only settings documented in the official material or implementation checks relevant to this feature and the primary task.
| Configuration item | Hypothetical trial starting point | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge base | Associate only the base containing A-01 | Check the vector model identifier of each base first and exclude those on a different model |
| Search mode | Hybrid retrieval | Compare recalled passages against semantic retrieval and full-text retrieval, observing identifiers and natural-language questions |
| Citation limit | Start from returning the full text of A-01 | Compare with a smaller value and observe whether the passage is truncated |
| Minimum relevance score | Start from a value that keeps A-01 and excludes unrelated passages | Compare with a higher value and observe whether A-01 is filtered out |
| Question optimization | Enable for follow-up questions containing references | Compare the original question with the optimized question and check whether the subject was altered |
| Result reranking | Enable according to settings | Compare the ranking position of A-01 before and after reranking |
Three common errors
- Treating pending review as confirmed. The input is the A-01 source text and the draft sentence "has confirmed that the Hangzhou warehouse can arrange 120 people per day from October 20." The comparison is between the status field "pending review" and the draft sentence's "confirmed." If the two take mutually exclusive values on the same field, the draft sentence is unsupported.
- Associating knowledge bases across vector models. The input is the vector model identifiers of two bases. The comparison is whether the two identifiers match. If the identifiers differ, the bases cannot be associated together, and a cross-model retrieval should not be run to compare missed recalls.
- Using the minimum relevance score as a permission filter. The input is the minimum relevance score setting and one unauthorized request. The comparison is whether that setting changes the authorization state of the request. The threshold affects relevance filtering only, and the authorization state stays unchanged.
How to verify the setup
- For the A-01 source text and the draft sentence, input that sentence and run retrieval, then compare whether the recalled passage contains the wording "pending review." If the draft sentence still says "confirmed," the sentence is unsupported and the evidence gap should be recorded.
- For empty-result handling, input the explicit paper input [] and compare whether the downstream design handles it as an empty citation. Do not claim that a fictional rule retrieval necessarily returns [].
- For cross-model association, input the vector model identifiers of two bases and compare whether the identifiers match. If they do not match, exclude that base and record the actual observation. Do not infer the runtime result of associating them together.
This page addresses only peak-sale warehouse and delivery scheduling, cross-border material and multilingual retrieval, and knowledge base retrieval and recall. At operational handover, which materials, operations, and requests sit in which authorization state must not be treated as retrieval or a form executing authorization.