What the data for this category looks like
Daily documents for this category include gameplay design docs, numerical tables, version requirement sheets, and balance test records. Data sources include player behavior data, public competitor materials, battle logs, and version feedback. Update frequency is mostly weekly or per version cycle. Most documents contain structured numerical fields and rule description text. Some numerical tables have clear units such as damage multiplier and experience required for leveling up. Data must be updated quickly alongside version iterations.
What constraints these characteristics impose on knowledge base retrieval and recall
The mixed document structure of structured numerical tables and rule text requires retrieval to support both semantic matching and exact field matching. Frequently updated version data requires the retrieval pipeline to support fast knowledge base refreshes. Multi-source player and battle log data requires separate retrieval permissions for public materials and internal test data. Retrieval results for numerical content must maintain precision to avoid rule deviations caused by fuzzy matching. The timeliness of version feedback requires recalled results to prioritize documents from the latest version.
How to configure settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Aligns with the precision requirements for numerical rules and structured documents, avoids irrelevant low-match content interfering with retrieval |
limit | Top 8–12 entries | Controls the total amount of content retrieved per call, adapts to the context load of long numerical tables and rule text in design documents |
searchMode | Hybrid semantic and exact matching | Covers both semantic queries for gameplay logic and exact retrieval needs for numerical fields |
usingReRank | Enabled | Re-ranks initial recall results to prioritize content related to current version rules and numerical design |
datasetSearchUsingExtensionQuery | Enabled | Automatically optimizes search keywords for technical terms, adapts to the specialized terminology used in game design |
collectionIds | Specify the knowledge base collection ID for the corresponding game project | Isolates document data across different game versions, avoids invalid retrieval across projects |
Three common configuration mistakes
- Symptom: Uploaded numerical table documents fail to chunk correctly, with long text truncated or fields split incorrectly. Cause: Chunking rules are not adjusted for structured table documents; the default chunking logic cannot recognize the row and column structure of tables.
- Symptom: Image understanding models cannot be configured when creating a new knowledge base, causing design documents containing battle charts to fail parsing. Cause: The corresponding model plugin is not enabled during deployment, or the current version does not include image understanding capabilities.
- Symptom: Retrieval results include outdated gameplay documents from non-current versions, failing to match the latest version requirements. Cause: The
collectionIdsparameter is not specified, so the retrieval scope covers all knowledge base collections.
How to confirm correct configuration
- Upload a typical numerical table document and run a chunking test. Confirm that the row and column structure of the table is correctly recognized, with no truncation or incorrect splitting.
- Enter a query containing specialized game design terminology. Check the relevance of retrieval results to confirm that the
usingReRankanddatasetSearchUsingExtensionQueryconfigurations are active. - Switch between different version knowledge base collections to verify that the
collectionIdsparameter correctly limits the retrieval scope, only returning documents from the target version. - Conduct tests using the FastGPT v4.17.0 platform. Adjust the value range of the
similarityparameter, test the number of retrieval results across different thresholds, and confirm that matching precision meets current business requirements.
The values provided here are common starting points, and should be evaluated against your own specific 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.