What data for this category looks like
Daily documents for this category include four structured file types: classification assessment reports, gap analysis reports, rectification plans, and official assessment reports. Data is collected from results exported by assessment tools, on-site interview records, system inventories, and security device logs. Data is archived and updated per individual project, with an update frequency of once per project per year. Each project’s documents include fields such as project number, assessment target, control item list, and rectification records. No unified fixed unit exists; all business data is linked only at the project level.
Constraints Imposed by Category Characteristics on Context and Token Management
Data for this category originates from dispersed sources and is archived per project. Multiple source documents must be integrated to form context, so the context window must accommodate the total token volume of all documents for a single project to avoid token overflow. Single-project data has a low update frequency but large single-load size. A balance must be maintained between the completeness of context recall and token consumption efficiency. The structured format of daily documents requires context to accurately match keywords tied to assessment stages, and filter out irrelevant content. The multi-stage connected nature of the assessment workflow requires context to retain historical data across stages. Incorrect configuration leads to incomplete recall of assessment materials, which can cause rectification delays or regulatory inspection risks.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Max Context | 8000–12000 token | Accommodates the total token size of the four document types per single project, prevents model context overflow |
| Max Reference Token | 2000–3000 characters | Matches the length of key paragraphs from a single core assessment document, ensures complete recall of business information |
| Number of Context Entries Recalled Per Single Request | Top 5 | Covers documents from the three core stages: classification assessment, gap analysis, and rectification, avoids redundant content consuming token quota |
| Context Trigger Rule | Based on current assessment stage tag | Accurately recalls historical data related to the current stage, reduces the proportion of invalid context |
| Global Context Storage | Enable project-level storage | Retains historical session data across stages, adapts to context requirements for multi-round assessment processes |
Three Common Configuration Errors
- Invalid JSON errors occur, with returned results containing unescaped control characters. This happens when special symbols such as line breaks and tabs in assessment reports are not preprocessed, resulting in invalid JSON format when context is concatenated.
- Context cannot be maintained across stages during multi-round conversations. This occurs when global context storage is not enabled, and only temporary context for a single round of conversation is retained. Historical data from different stages such as classification assessment, gap analysis, and rectification cannot be associated.
- Redundant content is still returned after the number of context recall entries exceeds the preset limit. This happens when automatic context truncation rules are not configured, causing excessive non-core documents to consume token quota and leading to model response timeouts.
How to Validate Proper Configuration
- Import all assessment documents for a single project, trigger context recall, and confirm that the number of returned context entries matches the preset number of single-round recall entries.
- Construct test text containing special symbols such as line breaks and tabs, submit the text, and confirm that the returned JSON format is valid with no unescaped control characters.
- Initiate consecutive multi-round queries associated with different assessment stages, and confirm that context retains key information from historical stages.
- Review system token consumption statistics, and confirm that token consumption for a single-project session does not exceed the preset maximum context threshold.
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.