Skip to content

RXDB Extractor

RXDB Extractor is a provenance-first tool for recovering relational records from hierarchical RedatamX/RXDB databases through supported RedEngine query semantics.

It is designed around a narrow promise: preserve source hierarchy and identity explicitly while producing relational Parquet datasets with inspectable provenance and validation evidence.

Choose your path

  • :material-rocket-launch: Run your first extraction

    Install the package, connect a RedEngine bridge, inspect a real database, qualify one bounded selection, then scale deliberately.

    :octicons-arrow-right-24: Getting started

  • :material-family-tree: Understand the data model

    Learn why hierarchical RXDB records need explicit generated identity before they can become safe relational tables.

    :octicons-arrow-right-24: RXDB and hierarchical data

  • :material-cog-outline: Review the mechanism

    Follow the software boundaries and the extraction path from inspection and profiles through FREQ normalization, key projection, validation, and artifacts.

    :octicons-arrow-right-24: Architecture

  • :material-shield-check-outline: Audit correctness

    See exactly what count, primary-key, foreign-key, extraction-invariant, checkpoint, and artifact checks establish—and what they do not.

    :octicons-arrow-right-24: Validation

  • :material-database-plus-outline: Adapt another RXDB source

    Treat a new database as an explicit adapter and qualification exercise instead of adding hidden source-specific behavior to the core.

    :octicons-arrow-right-24: Extending RXDB Extractor

  • :material-tools: Operate long runs

    Use provenance-aware resume, bounded workers, persistent bridge transport, and symptom-driven recovery without weakening the evidence contract.

    :octicons-arrow-right-24: Operations and troubleshooting

Implemented production path

flowchart LR
    A[RXDB hierarchy] --> B[Inspect]
    B --> C[Compile profile]
    C --> D[SELECTION + NUMBER + FREQ]
    D --> E[Normalize + merge by explicit IDs]
    E --> F[Project PK / FK]
    F --> G[Validate]
    G --> H[Parquet + provenance evidence]

The maintained production backend uses SELECTION + NUMBER + direct ancestor IDs + FREQ, with bounded variable batches and explicit-key merges. It does not depend on row position, TABLE VIEW, binary patching, or direct RBFX decoding.

The runtime boundary remains explicit: RedEngine is accessed through a protocol-v1 bridge and is not bundled by this Python package.

Reference implementation and evidence

Argentina Census 2022 is the first source-specific reference use case and is governed separately in the argentina-censo2022-rxdb repository. Database-specific source semantics, profiles, qualification evidence, and release decisions belong there rather than in the generic core.

For repository-level implementation status and preserved historical qualification results, see the project README.

Documentation map

The site separates four kinds of information:

The documentation truth audit is retained under Project internals for maintainers who need the evidence map behind the public documentation.