Bringing AI Answer Monitoring Data into the Data Warehouse: ODS, DWD, and ADS Three-Layer Design and Field Design

Bringing AI answer monitoring data into the data warehouse requires only three layers: ODS, DWD, and ADS. ODS only appends and stores raw answers; DWD performs normalization, deduplication, and entity tagging, and backfills by partition overwrite; ADS stores metric aggregations and must simultaneously record the numerator, denominator, and metric-definition version. The partition keys are selected as collection date plus batch, collection date plus platform, and metric-definition version plus statistical date, respectively. When the metric definition changes, the old version is frozen and the new version is replayed; ADS numbers are not allowed to be manually modified.