What the data for this category looks like
Data for new energy development documents and materials comes from official approval documents including project establishment approval, environmental impact assessment, grid connection acceptance, land pre-examination, and internal project execution documents. Update cycles follow project phase progression. Core documents such as feasibility study reports and grid connection plans receive updates at their corresponding approval nodes. Most documents use chapter-based structured content, with fixed fields including project number, land area, installed capacity, and approval document number. Units adhere to industry standard metrics including hectares, megawatts, and kilowatt-hours.
What constraints do these characteristics impose on the citation sourcing and traceability link?
The characteristics of new energy development document data create multiple constraints for citation sourcing and traceability. First, core data comes from official approval documents. Traceability requires precise matching of specific approval numbers and issuing entities, and cannot rely solely on vague source labels. Second, documents are updated in phases as projects progress. Traceability must verify the version timeliness of cited content, and avoid using outdated feasibility study or acceptance materials. Third, documents contain fixed professional fields and measurement units. Traceability must clearly mark the chapter position corresponding to each field, to ensure the traceability of cited content. Fourth, some documents include technical calculation bases. Traceability must link both data sources and derivation processes, to ensure citation compliance.
How to configure the setup
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale for This Setting |
|---|---|---|
| Similarity threshold | 0.75–0.85 | New energy development documents have dense professional terminology. This range filters low-relevance non-professional content, and balances recall accuracy and coverage |
| Single search limit | Top 8 results | Core reference information for new energy project documents is mostly concentrated in the first few recall results. 8 results cover main content and avoid redundancy |
| Reranking | Enabled | Semantic similarity of professional documents can be disrupted by terminology. Reranking corrects the order of search results, and improves matching of core fragments |
quoteMaxToken | 1200–1500 token | Valid content per segment of new energy documents is relatively long. This range ensures citation fragments contain complete professional expressions and calculation bases |
detail | Enabled | Full display of chapter and field information for cited content is required for subsequent traceability verification |
responseData | Return only citation fragments and sources | Focus on precise content needed for document drafting, and avoid returning irrelevant information that disrupts the drafting process |
Three common configuration errors
- Enabling reranking results in empty citation fragments. The cause is that reranking filters all results within the similarity threshold range, resulting in no eligible search fragments returned.
- Setting
quoteMaxTokentoo small, leading to truncated citation fragments that cannot fully display professional calculation bases and field information in new energy documents. - Failing to verify document versions, resulting in citations of outdated feasibility study reports. The symptom is that the approval number marked in the traceability does not match the latest approval document for the current project phase.
How to confirm the configuration is correct
- Upload an environmental impact assessment report for a new energy project, submit a request to draft relevant documents, and check if the returned results mark specific approval numbers and chapter positions.
- Adjust the reranking switch, compare the order changes of two search results, and confirm that the reranking function works properly.
- Check the length of citation fragments, confirm they fall within the range set by
quoteMaxTokenwith no truncation. - Test different similarity threshold values, observe changes in the number of recall results, and confirm that the threshold configuration works as expected.
The values listed above are common starting points, and should be evaluated against samples relevant to the actual deployment context.
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.