Compatibility And Package Status#

design-research is an optional umbrella package, not the only supported way to use the component libraries. This page records the exact package family tested by the current umbrella release and separates those compatibility facts from broader maturity policy.

Tested Package Combination#

Exact umbrella pins and current package metadata#

Package

Version

PyPI development classifier

design-research

0.4.0

Alpha

design-research-problems

0.4.0

Alpha

design-research-agents

0.6.0

Pre-Alpha

design-research-experiments

0.3.0

Alpha

design-research-analysis

0.3.1

Alpha

The versions match the exact dependencies in pyproject.toml. Normal python -m pip install design-research usage and pull-request CI resolve this published combination. The table does not promise compatibility for unlisted versions.

The status column reports each package’s current Development Status classifier; it does not define stable, experimental, research, or legacy states for the ecosystem. A shared maturity-label policy remains open in issue #12. Until that policy is decided, consult each package’s release notes and documentation for package-specific change guidance.

Artifact Contract#

The current deterministic family path writes an Experiments manifest with artifact schema 0.2.0. The umbrella family smoke test passes that exported artifact set to design-research-analysis==0.3.1 and validates its event table. This is the tested handoff:

Problems + Agents -> Experiments artifact set (schema 0.2.0) -> Analysis

Treat the artifact directory and its manifest schema as the cross-package data contract. A package version, API status, and artifact schema version answer different questions; none should be used as a substitute for the others.

Public API Scope#

The umbrella root exports __version__ plus problems, agents, experiments, and analysis. For the exact pins above, each wrapper mirrors its component package’s public __all__ exports. This provides one consistent import route for the tested family without claiming that every component symbol is permanently stable across future releases.

The no-network Canonical Artifact Flow exercises those public wrappers: it resolves a packaged problem, runs the public seeded baseline agent, exports canonical experiment artifacts, and validates them through top-level Analysis helpers. The live Prompt Strategy Comparison Walkthrough adds a prompt-built Agents workflow while retaining the same artifact handoff.

Start With The Umbrella Or Go Direct#

Choosing an install path#

Start with design-research

Install a component directly

You want one import route across Problems, Agents, Experiments, and Analysis for the tested combination.

You need only one package or a narrower dependency surface.

You want the shared learning path, composed examples, and compatibility record.

You need component-specific optional backends, internals, or detailed API guidance.

You are composing the complete artifact handoff.

You already know which package owns the behavior you need.

Direct component use remains supported by each owning package:

Next Step#

Start with Canonical Artifact Flow for the deterministic all-package handoff, then continue to Prompt Strategy Comparison Walkthrough only when you want the optional local-model runtime.