The loss of precision (e.g., 145677.755 appearing as 145677.00) following a migration to an Oracle Analytics Server (OAS) instance with a Databricks backend is a recognized challenge related to how the BI Server interprets numerical data types across different database drivers.

While individual column formulas can be adjusted, the sources identify several administrative and metadata-level configurations that can resolve this issue globally at the environment or Semantic Model (RPD) level.

1. Update the Physical Database Type and Features

In the Physical Layer of your semantic model, the BI Server uses the "Database Type" and "Features" settings to determine how to fetch and represent data from the backend.

• Verify Database Type: Ensure the Physical Database in the RPD has been updated from Netezza to the appropriate Databricks or Spark version supported by OAS. If a specific Databricks type is not available in your version of Model Administration Tool, using a generic "ODBC - Advanced" or "JDBC" type can sometimes resolve mapping errors.
• Numerical Feature Overrides: Navigate to the Features tab of the physical database object in the RPD. Check parameters such as CONVERT_NUMERIC_BITMASK or CONVERT_REAL_BITMASK. If these are incorrectly configured, the BI Server may assume the backend does not support high-precision floating points and truncate the values to integers during the fetch process.

2. Check for "Double Arithmetic" Blending

If your analysis combines a Subject Area with an external Dataset, the sources note that OAS may force the calculation to use binary floating-point approximations (DOUBLE arithmetic). This computing limitation can lead to precision errors where numbers are not represented exactly. If your measures are part of a blended visualization, verify if the "true" 0 or precise decimal is being lost during the join/blending process.

3. Leverage OAS System-Level Access (Global Fix)

Unlike Oracle Analytics Cloud (OAC), which is a restricted PaaS, OAS allows administrators to access and work on the system level, OS, and configuration files.

• Because OAS runs in the same manner as OBIEE 12c, you can inspect the NQSConfig.ini and DBFeatures.ini files.
• You can globally modify how the BI Server handles floating-point numbers or specific Databricks-specific data type mappings within the DBFeatures.ini for the corresponding driver to force the server to treat these columns as DECIMALor NUMERIC rather than INT.

4. Manual SQL Workaround

If global configuration changes do not resolve the issue, the documented workaround in the sources is to "manually convert a data column to a supported type by entering SQL commands". While you requested a global fix, wrapping your measures in a CAST function (e.g., CAST(Sales AS DOUBLE)) in the logical layer of the RPD is the standard method to force OAS to recognize the fractional part of the data if the underlying driver is misreporting the column precision.