What data for this category looks like
This category of data primarily originates from story scripts, game text, art resource descriptions during the game development phase, and publicly released regulatory compliance requirements. Daily maintained documents include content self-check lists, sensitive content modification records, user agreements, and compliance review minutes. Data is updated per version or event cycle. A single compliance review data entry includes fields such as text content, review status, modification trail, associated activity ID, and has no fixed unified format, adjusting dynamically alongside development progress.
Constraints on Context and Token Configuration
The long-text and high-frequency update characteristics of game content compliance self-check impose multiple constraints on context and token configuration. Story scripts and long-text content have high per-segment token usage, so total context token count must be strictly limited to avoid exceeding API call limits. Data updated per version or event requires the context to be associated with review records from the corresponding cycle, so the recall range must accurately filter irrelevant historical data. Different document types have significant differences in token density, so the token limit for single-document recall must be adjusted appropriately to prevent overall token overflow. Additionally, compliance self-check requires verifying API call costs, so the statistical dimensions of input and output tokens must be clearly distinguished.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
Maximum context token count (maxContext) | 8000–12000 token | Game story scripts have high per-segment length, need to cover multi-round review context while controlling API call costs |
Maximum single-document recall token count (quoteMaxToken) | 2000–3000 token | Game text and story scripts have high content density, prevent single documents from occupying too much context quota |
| Context recall count | Top 6 entries | Compliance review minutes are grouped by version or event, excessive recall will cause token overflow |
| Token statistics display | Enable separate input and output statistics | Aligns with compliance self-check cost accounting needs, facilitates checking quota usage |
| Context filtering threshold | Calibrated via actual testing | Differentiates recall priority between sensitive content and normal content, prevents irrelevant documents from occupying tokens |
Three Common Misconfigurations
- Only total token count is displayed in the interface, with no distinction between input and output tokens. Cause: The token statistics display configuration is not enabled, making it impossible to accurately verify API call costs for compliance self-check.
- Historical review records are still recalled after setting the context count to 0 in the workflow. Cause: Confuses the configuration logic between context recall count and historical conversation turns, fails to properly limit the context scope for compliance reviews.
- Markdown formatting in context references is not rendered correctly. Cause: Format markers were not retained when splitting knowledge base documents, or the
quoteMaxTokenconfiguration is too small, causing formatting fragments to be truncated.
How to Confirm the Configuration Is Correct
- View the AI call logs in FastGPT v4.17.0, confirm that input and output token counts are displayed separately, and check if the configuration takes effect.
- Upload a story script exceeding 2000 tokens, verify that the number of recalled context documents matches the set recall count.
- Test setting the context count to 0, confirm that no historical compliance review records are recalled.
- Upload a content self-check list with Markdown formatting, verify that the formatting displays normally in context references.
The values given 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.