Small, composable vocabularies and protocols for expressing analytical meaning, decisions, rationale, evidence, and conclusions.

composable-data provides a lightweight, typed grammar for analytical work.

The goal is to make important analytical structure explicit and interoperable without replacing the libraries that perform the analysis.

composable-data provides small shared vocabularies for expressing things such as:

• what one observation represents
• what kind of analytical problem is being addressed
• which features were selected and why
• how an experiment is split and why
• which model was chosen and why
• what evaluation evidence resulted
• what conclusion the evidence supports
• how a data-quality problem was resolved and why

It facilitates standard interactions with analytical tools that:

• load and transform data
• train models
• generate predictions
• calculate metrics
• create visualizations
• execute queries

Any additional packages should remain independently useful and composable.

The ecosystem follows this central rule:

Add a concept only when necessary to express a declaration that cannot otherwise be expressed correctly.

The shared vocabulary is intentionally small; it is not an analytics framework.

• Derive what can be known.
• Record what happened.
• Require rationale where alternatives require analyst judgment.
• Keep analytical computation in established libraries.
• Keep domain- and dataset-specific semantics outside the core grammar.
• Use established analytical language for public names.
• Keep packages small, inspectable, typed, and dependency-light.

• analytics-toolworks - reusable open-source tools for analytical work
• applied-models - reproducible model experiments on accessible real-world datasets

Together, these projects support a connected analytical workflow that enables analysts to work at a higher level:

composable-data
    shared analytical meaning and decision grammar
        ↓
analytics-toolworks
    reusable tools to inspect and communicate the work
        ↓
applied-models
    reproducible experiments on real data

Individual repositories specify their own licenses.