release: v0.5.0 grounded AI narratives

[suneelprojects]

Release summary

EazyDataFix v0.5.0 adds an optional, provider-neutral grounded narrative layer on top of completed deterministic Agentic EDA workflows. Deterministic metrics remain authoritative, existing non-AI workflows remain unchanged, and no LLM dependency or API key is required unless narrative generation is explicitly used.

Main user-facing features

• edf.generate_agentic_eda_narrative(...) for evidence-cited, business-facing narratives
• Provider-neutral NarrativeProvider interface
• Optional OpenAI Responses API adapter through the openai extra
• JSON-ready narrative, claim, evidence, and configuration models
• Optional grounded narratives in Agentic EDA HTML, Markdown, and JSON reports
• Public API and model exports for the narrative workflow

Grounding and integrity safeguards

• Providers receive an immutable compact evidence brief, not the raw dataset
• Every claim must cite supplied deterministic evidence
• Invalid, missing, duplicate, and unknown citations are rejected
• Invented numbers, unsupported causal claims, and insufficient lexical support are rejected deterministically
• A SHA-256 workflow fingerprint prevents attaching a narrative to a different workflow
• Narrative generation does not rerun or mutate the deterministic workflow or caller-owned DataFrames

These controls are deterministic guardrails, not proof of semantic truth; generated narratives still require human review.

Compatibility

• Package version: 0.5.0
• Python support: 3.10–3.13
• No existing public API was removed or renamed
• The OpenAI SDK remains an optional dependency
• Existing deterministic EDA, reporting, notebook export, and approval checkpoint behavior is preserved
• develop is 10 commits ahead and 0 commits behind main

Release validation

• python -m pytest: 194 passed, 1 skipped (195 collected)
• ruff check .: passed
• black --check .: passed (177 files unchanged)
• git diff --check origin/main...origin/develop: passed
• Merge-conflict marker scan: passed
• python -m build: passed
  • eazydatafix-0.5.0-py3-none-any.whl
  • eazydatafix-0.5.0.tar.gz
• python -m twine check dist/*: passed for all checked distributions, including both v0.5.0 artifacts
• Wheel metadata: Version 0.5.0, Requires-Python: >=3.10
• Clean-wheel installation: passed in Python 3.13.9 with pandas 3.0.5
• Clean-environment import: eazydatafix.__version__ == "0.5.0"
• Grounded-narrative public imports: passed
• Local/remote v0.5.0 tag check: not present
• PyPI check: latest published EazyDataFix version is 0.4.0; v0.5.0 is not published
• Release comparison review: only intended v0.5.0 implementation, tests, public exports, reports, documentation, changelog, release notes, and version metadata are included; no secrets, generated build artifacts, temporary files, or local configuration files are tracked

Post-merge release checklist

• Create and push the v0.5.0 tag
• Create the GitHub Release from the tag
• Publish the validated artifacts to PyPI
• Verify the published package and release links

Tagging, GitHub Release creation, and PyPI publication intentionally remain pending until this pull request is reviewed and merged.