- **Description**: A specialized index type (provided by the `bloom` extension) that implement a **Probabilistic Data Structure**. It is designed for tables where queries involve arbitrary combinations of many different columns, but only require equality checks. - **Physical Implementation**: - **Bit Signatures**: For each heap row, a fixed-length **Signature** (bit-vector) is generated. Each indexed column is hashed multiple times to set specific bits in this signature. - **The Page Format**: Signatures are stored sequentially in standard 8KB pages. Unlike B-Tree, there is no tree structure; the entire index is a series of flat signature pages. - **Signature Length**: Configurable via `signature_size`. A larger signature reduces the "False Positive" rate but increases the index size. - **Query Logic**: - **Signatures**: Each row's indexed columns are hashed into a fixed-length bit-signature (e.g., 80 bits). These signatures are packed into standard 8KB pages. - **The Filter Scan**: Postgres performs a sequential scan of the **Index** (not the table), performing a bitwise **AND** between the query signature and each stored row signature. - If `(RowSignature & QuerySignature) == QuerySignature`, it is a potential match. - **Supported Operators**: `=`. - **Special Features**: - **Space Efficiency**: Can be significantly smaller than multiple B-Trees or a single wide GIN/GiST index. - **Probabilistic**: Can return **False Positives** (signatures match by coincidence). Postgres handles this by always performing a "Recheck" against the actual heap tuple. It never returns "False Negatives."