What the documents look like
Cross-border platform settlement and foreign-exchange work concerns the collection of funds from orders that have already been completed. It does not cover fraud chargebacks. The main data sources are the platform settlement center, bank statements, payment institution back ends, and ERP orders. The documents handled day to day are platform settlement statements, exchange-rate tables, service-fee invoices, and receipt reconciliation tables. Updates arrive weekly or monthly, and retrieval is scoped to the current week or month settlement batch rather than to a single transaction.
For copy drafting in this category, each proposed sentence must be checked against the wording that actually appears in a settlement statement, exchange-rate table, service-fee invoice, or reconciliation table. Any rate, settlement cycle, currency, or service-fee basis that appears in the draft needs a corresponding field or sentence in those materials. A figure or commitment that the materials do not state remains unsupported.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. The smallest search unit is the knowledge base; a collection is only a way to organize data and does not affect search results. Semantic search relies on vector distance. Full-text search suits precise lookups such as numbers, model names, and proper nouns. Hybrid search combines the two paths through RRF. The minimum relevance value takes effect only in semantic search or when result reranking is used, and it acts as a filter threshold. The reference limit is measured in tokens and caps how much content a single search may cite. Question optimization performs coreference resolution and question expansion and adds one model call.
Retrieval does not handle permissions or entity isolation. The minimum relevance value is a relevance filter and does not exclude records by business entity or authorization scope. Knowledge bases that use different vector models cannot be associated together. During checking, first inspect the vector model identifier, exclude any that do not match, and record what is observed. Do not run a cross-model search to compare missed recalls.
Hypothetical paper exercise: A cross-border seller's finance staff member prepares a product copy line reading, "This store's settlement cycle is every Friday, settled in US dollars, with the service fee deducted according to the amount listed on the settlement statement." The materials provided for checking are: platform settlement statement (settlement batch 2024-W21), which states "Settlement cycle: every Friday; settlement currency: USD; service fee: deducted according to the amount in the fee column of the settlement statement." Exchange-rate table (May 2024), which states "USD/CNY mid-rate 7.10." Service-fee invoice (number SF-0521), which states "Service fee amount is subject to the fee column of the settlement statement." The sentence to be judged is "This store's settlement cycle is every Friday, settled in US dollars, with the service fee deducted according to the amount listed on the settlement statement." The scope of judgment is limited to settlement batch 2024-W21, currency USD, and the fee column field.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this page | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Platform settlement statement base, exchange-rate table base, service-fee invoice base | The three bases must use the same vector model; first check the vector model identifier and exclude any that do not match |
| Search mode | Hybrid search | The draft sentence contains exact terms such as "every Friday" and "USD" and also natural-language phrasing; compare with pure semantic and pure full-text results |
| Reference limit | Hypothetical trial starting point of 2000 tokens for this exercise | The settlement statement fee column and invoice number need to be quoted in full; compare with 1500 and 2500 to see whether content is truncated |
| Minimum relevance | Hypothetical trial starting point of 0.3 for this exercise | Takes effect only in semantic search or reranking; compare with 0.2 and 0.4 to see whether irrelevant fragments are filtered out |
| Question optimization | Enabled | The draft sentence contains the reference "this store"; compare the original question with the optimized question to check whether the subject is altered |
| Result reranking | Enabled | Candidates include multiple paths from the settlement statement, exchange-rate table, and invoice; compare scores and ordering before and after reranking |
The values above are hypothetical trial starting points for this exercise and are not official default values. The comparison method for the reference limit and minimum relevance is to test content completeness with actual materials of different lengths, then adjust the values.
Three common errors
- Treating minimum relevance as a permission or entity filter. Input the draft sentence and the three knowledge bases, then compare whether returned results contain records from other settlement batches or other currencies. Minimum relevance filters low-relevance fragments by a 0-1 score and does not exclude records by business entity. If returned results contain content outside batch 2024-W21, that is a retrieval scope issue and requires checking the associated knowledge base identifiers; raising the relevance value does not resolve it.
- Treating a pending-review status as approved. Input service-fee invoice number SF-0521 and compare its status field. If the invoice shows only "pending review," that status does not show the fee has been approved for deduction, and it does not show a lack of operating permission. The draft sentence's "deducted according to the amount listed on the settlement statement" must rely on the original wording in the settlement statement fee column; the invoice review status is a separate fact.
- Associating knowledge bases that use different vector models. Input the platform settlement statement base and another exchange-rate table base that uses a different vector model, then compare the two vector model identifiers. When the identifiers do not match, they cannot be associated together. Record what is observed; do not infer missed recalls or redundant returns as an operating result. First check the identifiers and exclude those that do not match.
How to verify the setup
- Input the draft sentence from the hypothetical paper exercise and the three knowledge bases, then compare whether "every Friday," "USD," and "service fee deducted according to the amount listed on the settlement statement" are each supported by the settlement cycle field, the settlement currency field, and the original wording of the fee column in the settlement statement. The expected observation is that all three have corresponding original text; if any lacks a corresponding field or sentence, record it as unsupported.
- Input the retrieval terms before and after question optimization, then compare whether the optimized question retains the settlement batch 2024-W21 subject referred to by "this store." The expected observation is that the subject has not been replaced with another batch or another store; if the subject is altered, record the question optimization result and check the original input.
- Input an empty-result scenario and use the explicit paper input [] to check the downstream design. The expected observation is that the cited content is output as an array with length 0 and the output path completes. Do not claim that a fictional rule search necessarily returns []; only check whether the downstream receives an empty array as designed.
The verification above is paper checking. It does not claim that testing, deployment, or approval has occurred. Missing fields remain missing, pending review does not constitute approval, and a quotation validity period does not indicate a supply period.