What Data for This Category Looks Like
This category of data typically originates from procurement systems, supplier quotation documents, budget sheets, and contract systems. Daily processed documents include purchase requisitions, price comparison records, formal contracts, and acceptance documents. Data is updated per order or monthly. Each individual document includes fields such as purchase order number, supplier qualification information, budget amount, and clause details. There is no unified fixed format: some documents are structured tables, while others are unstructured text.
What Constraints These Characteristics Impose on Context and Token Processing
Procurement-related documents have wide variation in single-document length. Short purchase requisitions are only a few hundred characters, while long contract clauses can reach several thousand characters. The per-order update feature means context must bind all associated documents for the corresponding procurement batch. Mixed multi-source data introduces differences in field formats, so invalid fields must be filtered during context splicing to avoid wasted token usage. Bulk monthly updated data can cause total context token volume to exceed thresholds, so the scope of documents associated with a single review round must be limited to prevent exceeding the model’s token limit. Additionally, change review requires associating historical procurement records with current applications, so context must accurately match past price comparison data for the corresponding supplier to avoid irrelevant information interfering with valid token payloads.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 15000–20000 token | A single review round for this category requires associating three document types: purchase requisitions, price comparison records, and historical contracts. Total token volume typically falls within this range |
| Single Recall Limit | Top 4 associated documents | Procurement data is updated per order, and associated documents per batch do not exceed 4. Exceeding this number will exceed the token limit |
| Chunk Length | 800–1200 characters | Procurement contract clauses are usually coherent paragraphs. Too long chunks reduce retrieval accuracy, while too short chunks destroy clause integrity |
quoteMaxToken | 3000–4000 token | Change review requires referencing core clauses. Valid token volume for a single core document typically falls within this range |
| Context Splicing Rule | Bind associated documents by procurement batch | Procurement data is managed per order. Cross-batch context introduces invalid information that interferes with review judgments |
Three Common Misconfigurations
- Phenomenon: Attempting to pass custom context via the API but it does not take effect. Cause: The system historical context splicing function was not disabled, resulting in duplicate splicing of custom context and system historical content, which exceeds the token limit.
- Phenomenon: Segmented procurement documents fail to accurately match corresponding business clauses. Cause: The chunk length was set too small, destroying the integrity of coherent business content such as contract clauses and price comparison details, leading to context association failure.
- Phenomenon: Change review classification results always fall into the fallback category. Cause: No configuration was made to associate historical context by procurement batch. Only single pieces of information from the current question are used for judgment, making it impossible to match the past price comparison rules of the corresponding supplier.
How to Verify Proper Configuration
- Import a single purchase requisition along with its associated price comparison records and historical contracts, then check the context preview module to confirm that only documents from the current batch are associated.
- Trigger a change review task, then check the token consumption log to confirm that the actual consumption value does not exceed the configured
maxContextlimit. - Adjust the chunk length parameter, import a long contract document, and check whether the chunked results retain complete clause paragraphs.
- Check the
quoteMaxTokenconfiguration to confirm that the token consumption of core referenced documents does not exceed the set range.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.