What data for this category looks like
The daily documents for this category include gameplay design documents, numerical tables, version requirement sheets, and balance test records. Data sources mostly include player behavior data, publicly available competitor materials, battle logs, and version feedback. The core fields of document structure are rule descriptions, numerical parameters, and test results. Numerical tables often include units such as percentages and fixed damage values. The data update rhythm follows weekly or per-version iterations, with content changing dynamically as versions are adjusted and test results are updated.
What constraints do these characteristics impose on knowledge base retrieval and recall
Documents in this category combine structured numerical parameters and semi-structured rule descriptions, and are updated frequently with version iterations. This requires retrieval to accurately match target document content and avoid retrieving irrelevant cross-version data. Internal ticket and on-duty scenarios mostly focus on current version rules and numerical issues. If the retrieval range is too large, it is easy to mix in old version test records or irrelevant competitor materials, leading to deviations in subsequent rule breakdown and numerical simulation, ultimately causing gameplay imbalance or version rework. At the same time, differences in units and fields within documents require retrieval to match corresponding dimension parameters, avoiding invalid cross-unit retrieval.
How to configure settings
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match precise parameter descriptions in numerical tables and design documents, filter low-relevance cross-version documents |
limit | 200–500 items | Cover the number of core documents within a single version, avoid excessive single retrieval causing context overflow |
searchMode | Precise matching | Prioritize retrieving highly relevant content for structured numerical fields and rule descriptions |
usingReRank | Enabled | Re-rank initial retrieval results, prioritize content associated with the current version |
datasetSearchUsingExtensionQuery | Disabled | Prevent automatic query expansion from retrieving irrelevant competitor or old version materials |
collectionIds | Bind dedicated knowledge base collections for the current version | Isolate cross-version historical documents, ensure only content within the current iteration range is retrieved |
Three Common Misconfigurations
- Phenomenon: A large number of non-target documents are returned during retrieval, including old version test records and competitor materials. Cause: The dedicated
collectionIdsfor the current version are not bound, so the retrieval range covers the full knowledge base. - Phenomenon: Only two preset options (100 or 900) are available when configuring
limit, and custom values around 300 cannot be entered. Cause: The interface dropdown menu does not allow custom input within the range, only preset options are supported. - Phenomenon: A large number of low-relevance reference content is still returned even after setting
similarityto 1. Cause:usingReRankis not enabled, so initial low-similarity retrieval results are not filtered and directly enter the context.
How to Confirm Proper Configuration
- Verify that the currently bound
collectionIdsonly include documents such as current version gameplay design documents and numerical tables. - Input precise numerical parameters from a single document for testing, confirm that retrieval results only come from target documents.
- Check that
usingReRankis enabled, and confirm that the re-ranked results prioritize current version content. - Verify that the configuration interface parameters for FastGPT v4.17.0 match the preset configuration.
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.