What this category of data looks like
Declaration materials and case review data for refinery production mainly originate from unit operation logs, compliance declaration dossiers, quality inspection reports, and safety review documents. Data updates follow individual declaration batches or monthly schedules. Each document contains structured fields and long-text review notes. Structured fields include unit number, process pressure, approval document number, and other items. Units involve industrial parameters such as MPa, ℃, and tons. Case documents are mostly multi-page structured combinations, with the longest single document reaching thousands of characters.
Constraints on Source Citation and Traceability
The coexistence of structured fields and long text requires precise matching of field content and long-text fragments during traceability, to avoid confusing process parameters and review notes. The batch-based update rhythm requires traceability logic to bind the unique identifier of the declaration batch, to avoid recalling irrelevant dossiers across batches. The coexistence of multiple units in fields requires synchronous marking of corresponding units when displaying citations, to allow reviewers to quickly identify parameter validity. Documents with a high proportion of long text require retaining context association during recall, to avoid truncating key review logic.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 1500–2500 token | Refinery declaration materials have long individual texts, requiring coverage of sufficient review details to avoid truncating critical content |
responseData | Enabled | Source traceability dossiers and declaration material entries must be displayed in results to meet compliance requirements for review traceability |
detail | Enabled | Refinery dossiers contain multi-field information, requiring display of complete document fragments and field content |
| Single recall count | Top 6 entries | Associated documents for a single batch of declaration materials are concentrated between 5 and 8 copies, avoiding redundant recall that disrupts review |
| Context window | 3000–4000 characters | The preceding and subsequent logic of declaration materials must be retained to ensure coherence of review judgments |
| Citation identifier binding | Approval document number | The approval document number of refinery dossiers is a unique identifier, enabling precise association of original documents with corresponding declaration batches |
Three Common Configuration Errors
- No cited dossier entries are displayed in conversation results, and no
responseDatarelated return fields are included in the conversation interface call. The cause is that theresponseDataconfiguration item is not enabled, resulting in traceability information not being included in the return fields. - Recalled citation documents do not match the current declaration batch, with dossier content from cross-batches appearing. The cause is that the approval document number is not bound as a unique identifier, and the recall logic does not filter dossier data for the corresponding batch.
- Citation fragments are too long, exceeding the scope required for review, and irrelevant historical process data is introduced. The cause is that
quoteMaxTokenis set too high, failing to limit the recalled text length.
How to Verify Successful Configuration
- In the FastGPT v4.17.0 environment, initiate an audit query for a specific batch of refinery declaration materials, and check whether the returned results include associated dossier entries and source identifiers.
- Enter the knowledge base configuration page, check the active status of parameters such as
quoteMaxToken,detail, andresponseData, and confirm they match the preset values. - Submit a new refinery declaration dossier for incremental update, initiate a query, and confirm that the newly submitted document can be correctly recalled and associated.
- View the conversation log, and confirm that the returned
responseDatafield contains complete cited document information and approval document number identifiers.
The values listed here are common starting points, and should be evaluated against the reader's own specific samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.