Package Surface

extended-data is a base Python data package. Its public contract is the extended_data namespace plus the console command extended-data.

Tier 1 Primitives

Pure functions live under extended_data.primitives. These functions cover structured serialization, scalar coercion, matching, key normalization, sequence/mapping utilities, state predicates, stack inspection, and redaction.

from extended_data.primitives import decode_json, encode_yaml, redact_sensitive_text

payload = decode_json('{"service": "api"}')

assert payload == {"service": "api"}
assert "service: api" in encode_yaml(payload)
assert redact_sensitive_text("token=abc") == "token=[REDACTED]"

Tier 1 names are intentionally not exported from the package root. Import them from extended_data.primitives so the root remains reserved for cohesive data surfaces.

Tier 2 Containers

Extended containers promote decoded data into ergonomic objects:

  • ExtendedData

  • ExtendedString

  • ExtendedDict

  • ExtendedList

  • ExtendedTuple

  • ExtendedSet

from extended_data import ExtendedData, ExtendedDict, ExtendedList, ExtendedString

assert ExtendedString("api-gateway").to_snake_case() == "api_gateway"
assert ExtendedList([None, "", {"service": "api"}]).first_non_empty()["service"] == "api"
assert ExtendedDict({"enabled": "true"}).reconstruct_special_types()["enabled"] is True
assert ExtendedData({"service": "api"}).merge({"replicas": 2}).as_builtin()["replicas"] == 2

ExtendedData is the generic facade for any incoming value. Shape-specific containers keep Python collection behavior while adding methods that route through Tier 1 primitives.

Tier 3 Processors

Tier 3 surfaces compose primitives and containers for real data movement:

  • DataFile reads, decodes, tracks metadata, and exports structured files.

  • DataWorkflow layers file reads, merges, transforms, writes, syncs, and provenance.

  • InputProvider loads direct inputs and environment data.

  • Logging handles structured lifecycle logging and returns stored snapshots as extended containers.

from extended_data import DataFile, DataWorkflow, InputProvider, Logging

artifact = DataFile.decode("service:\n  name: api\n", suffix="yaml")
workflow = DataWorkflow.from_value(artifact.data).merge({"replicas": 2}).result()
inputs = InputProvider(inputs={"ENV": "dev"}, from_environment=False)
logger = Logging(logger_name="docs", enable_console=False, enable_file=False)

logger.logged_statement("workflow ready", json_data=workflow.as_builtin(), log_level="info")

assert artifact.metadata["encoding"] == "yaml"
assert workflow.as_builtin()["replicas"] == 2
assert inputs.inputs["ENV"] == "dev"

CLI

The extended-data command exposes the same file and workflow boundary:

extended-data decode '{"service": {"name": "api"}}' --suffix json
extended-data decode --file config.yaml --output json
extended-data inspect --file config.yaml
extended-data merge config/base.yaml config/dev.yaml --output yaml
extended-data transform --file payload.json --step reconstruct --step unhump

Pytest Plugin

pytest-extended-data is a separate developer-facing distribution in the same workspace. It publishes pytest fixtures and assertion helpers through the standard pytest11 plugin entry point, so downstream packages can opt into Extended Data test helpers without installing pytest behavior from the runtime package.

uv add --dev pytest-extended-data

def test_payload_shape(extended_data_value):
    assert extended_data_value.shape == "mapping"
    assert extended_data_value["service"]["name"] == "api"

Split Packages

vendor-fabric owns external API clients, optional provider SDK dependencies, provider-specific examples, vendor capability dispatch, and vendor-backed Python sync.

agentic-fabric owns agent orchestration, runtime selection, agent-facing tool catalogs, MCP agent adapters, and framework adapters such as LangChain, CrewAI, LangGraph, and Strands.

extended-data does not preserve extended_data.connectors, extended_data.secrets, extended_data_types, directed_inputs_class, or lifecyclelogging compatibility shims. Those import failures are intentional in this major version.