Citation Sources and Traceability for Steel Trade Intelligent Due Diligence Reports

Steel trade data sources include public transaction data from domestic bulk commodity spot trading platforms, steel mill ex-factory quotation systems

What the data for this category looks like

Steel trade data sources include public transaction data from domestic bulk commodity spot trading platforms, steel mill ex-factory quotation systems, customs import and export clearance data, and industry association supply and demand reports. In terms of update frequency, spot trading data updates hourly, ex-factory quotations update daily, clearance data updates per customs declaration batch, and industry reports are released monthly. Documents are mostly structured CSV/Excel or semi-structured PDF reports, with fields including trading target name, specification model, origin, unit of measurement (ton), transaction unit price (yuan/ton), transaction volume, transaction date, trading entity name, and logistics origin/destination.

Hourly updates of spot trading data require that traceability must carry a timestamp precise to the hour as the traceability identifier, otherwise the corresponding batch of transaction records cannot be located. Field differences across multiple data sources: for example, some data sources include a "steel mill guidance price" field, while others include "actual transaction price". Traceability requires clear specification of field mapping rules to avoid confusion between the two types of prices. The fixed column format of structured documents requires locking a specified column range during parsing, otherwise irrelevant remark fields will be extracted. Batch association in customs data requires binding the customs declaration number as the unique traceability identifier during traceability, to ensure that referenced import and export data can be traced.

How to Configure

Configuration ItemRecommended ValueRationale
chunk_size800–1200 charactersThe combined length of core fields for steel trade transaction data is mostly 600-1000 characters. This segmentation range can fully retain key information such as specifications, prices, and entities of a single transaction record
chunk_overlap100–150 charactersAvoid breaking cross-field contextual associations during splitting, ensuring that associated logistics and payment information can be fully matched during traceability
recall_top_kTop 8–12 entriesDue diligence reports need to cover multi-dimensional data. This recall volume balances information richness and report readability
similarity_threshold0.75–0.85There are similar specification descriptions in the steel trade category. This threshold can filter low-match results while retaining valid data with similar semantics
SOURCE_TIMESTAMP_FIELDtransaction dateThe update of steel trade data is strongly bound to transaction time. Using this field as the traceability timestamp can accurately locate the original data batch
PARSE_FILE_FIELD_MAPPINGCalibrate based on actual testingThere are differences in field names across different data sources. Mapping rules need to be configured for exclusive fields such as transaction order number and transaction unit price to ensure unified fields after parsing

The parameter values provided on this page are conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require case-by-case analysis. It is recommended to test on your own samples before finalizing settings.

Three Common Mistakes

  • Phenomenon: A 400 Bad Request error appears when viewing knowledge base citations in a due diligence report. Cause: The SOURCE_TIMESTAMP_FIELD parameter is not configured, the system cannot generate a valid traceability identifier, and incomplete request parameters trigger verification failure.
  • Phenomenon: No citation content is displayed in the due diligence report. Cause: recall_top_k is incorrectly set to 0, no source data is recalled, and traceability information cannot be generated.
  • Phenomenon: An empty value is returned when referencing variables. Cause: PARSE_FILE_FIELD_MAPPING is not configured, the system cannot identify exclusive steel trade fields such as "transaction unit price" and "origin", resulting in empty extracted target fields.

How to Confirm Configuration is Complete

  • Upload a single steel trade transaction data document, check the parsed field list, and confirm that exclusive business fields have been correctly identified and extracted.
  • Initiate a due diligence query, check the traceability binding information of each returned content, and confirm that all carry the preset timestamp-based traceability identifier.
  • Modify the values of recall-related configurations, verify whether the number of query result recall entries changes correspondingly with the configuration adjustments.
  • Toggle the switch state of the citation traceability configuration, confirm whether citation annotations are displayed or hidden in the due diligence report, and verify that the configuration takes effect.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.