What This Category’s Data Looks Like
Data sources for this category include ERP material master data, melting furnaces and thermometers, spectrometers, MES heat records, mold management systems, customer drawings, and technical specifications. Daily generated documents cover foundry process cards, batching lists, heat records, mold histories, first article inspection records, material test reports, and nonconforming product review forms. Data updates occur per heat, per batch, or daily. Fields include alloy composition percentage, melting temperature, pouring parameters, mold cycle count, defect type, and more, with corresponding physical units attached.
Constraints Imposed on Source Citation and Traceability
Because data updates per heat or batch, citation and traceability must tie to unique heat or mold cycle identifiers to avoid cross-batch confusion. Multi-source heterogeneous data comes from equipment, systems, and drawings. Clear field mapping rules must be established for each source to ensure cited batching ratios, temperature parameters, and original records fully match. Professional fields such as alloy composition percentage and pouring temperature include units; original units must be retained to ensure traceability accuracy. Nonconforming product review forms are linked to heat and mold cycle numbers. Traceability must cover the full chain from defect to production link to avoid gaps. Frequently updated data must pull the latest version in real time to prevent use of expired heat records or batching lists.
Configuration Settings
The following configurations are adapted for FastGPT v4.17.0. Specific values can be adjusted based on actual business scenarios.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Maximum Context | 10000–14000 token | Adapts to the average single-document length of foundry-related materials, and accommodates stitching needs for multi-source associated data |
| Single Search Limit | Top 6–10 entries | Covers core associated documents such as process cards, heat records, and mold histories for a single heat |
quoteMaxToken | 2500–4500 token | Limits the character length of a single citation to avoid overly long fields disrupting context structure |
responseData | Enable original field retention | Ensures cited content includes unique identifiers such as heat numbers and mold cycle numbers to support precise traceability |
detail | Enable full metadata retention | Retains document update time and source system information to improve traceability across the full chain |
| Chunk Overlap Length | 100–150 characters | Retains associated fields across heats and mold cycles, preventing key information from being truncated during chunking |
Three Common Configuration Errors
- Phenomenon: The number of context entries displayed on the interface does not match the number sent to the caller. For example, the interface shows 30 entries but 310 are actually sent. Cause: The matching relationship between
quoteMaxTokenand maximum context is not correctly configured, and the system truncation logic is not aligned with front-end display logic. - Phenomenon: Document chunks exceeding the set citation limit are incorrectly cited. Cause: Chunk length settings do not cover splitting of ultra-long documents, or do not adapt to long-text processing logic, resulting in large chunks being directly truncated and cited.
- Phenomenon: The cited alloy batching ratio does not match the original heat record. Cause: Original field retention for
responseDatais not enabled, and system-automatically stitched fields lose the associated mapping between heat numbers and mold cycle numbers, leading to traceability errors.
How to Verify Correct Configuration
- Review the citation log to confirm each cited entry includes unique identifiers such as heat numbers and mold cycle numbers, and verify the enabled status of
responseData. - Test importing a single ultra-long foundry process card, check if chunked content retains key fields such as melting temperature and batching ratio, and verify the effect of the chunk overlap length setting.
- Initiate a quality audit query, compare the number of context entries displayed on the interface with parameters received by the caller, and verify the matching degree of maximum context and
quoteMaxTokenconfigurations. - View the metadata panel to confirm each cited document displays update time and source system information, and verify the enabled status of
detail.
The values provided are common starting points and should be evaluated against the reader's own samples.