Pillars

These pillars define what belongs in extended-data.

Pure Data First

Tier 1 primitives are plain Python functions. They should be deterministic, easy to test, and independent of files, logging state, vendors, agent runtimes, network clients, or environment configuration.

Real Python Containers

Tier 2 containers are real Python objects with the closest native behavior for their shape. ExtendedString, ExtendedDict, ExtendedList, ExtendedTuple, and ExtendedSet inherit from ExtendedData and from the appropriate Python base.

Promote On Boundaries

Data entering the package should be promoted into extended containers. Nested writes should keep promoting incoming built-in values. Data leaving the package should cross explicit boundaries such as as_builtin(), to_export_safe(), file writes, workflow results, or sync operations.

Compose Higher Layers

Tier 3 features compose Tier 1 and Tier 2. File processing, input loading, logging snapshots, imports, exports, and workflows should use shared data primitives instead of reimplementing merge, redaction, encoding, coercion, or promotion behavior.

Keep Vendors Out

Vendor clients, cloud SDKs, agent runtimes, and live API sync behavior do not belong in extended-data.

No Compatibility Shims

This package is a clean major-version surface. Removed package names and removed in-package connector namespaces should fail to import. Breakage is preferable to silently carrying stale contracts.

Docs And Tests Together

Public behavior must be backed by tests and docs in the same change. Examples are executable contracts. Sphinx/Furo docs, autodoc2 API output, README guidance, unit tests, and release hygiene checks all need to agree.