VSF Data Profiler v0.2.0-rc2

VSF Data Profiler v0.2.0-rc2

Date: 2026-06-16

This prerelease publishes the validated US-055 and US-056 increments on top of
v0.2.0-rc1.

Highlights

• Added canonical table_assessments.json with one deterministic assessment
  per profiled table: inferred role, health score, readiness, issue counts,
  affected columns, relationship risks, business-impact category, evidence
  references, and next actions.
• Surfaced table assessments in Markdown/HTML reports, export packages,
  artifact audit, web-runner artifact links, dashboard filters, and drilldowns
  without changing existing artifact names.
• Extended optional L4 narrative context and guardrails so table/business-impact
  claims must be supported by table_assessments.json.
• Redesigned the local web runner as a restrained data-quality console with a
  run-first first screen, clear 127.0.0.1 local runner boundary, post-run
  dashboard focus, and dedicated Table Impact section.
• Added static UI and Playwright coverage for the new console, Table Impact
  dashboard, artifact-driven graph/drilldown behavior, and no raw CSV dashboard
  reads.

Validation

Release validation in the product checkout:

node --check web/app.js
.venv/bin/pytest -q
.venv/bin/ruff check src tests scripts/verify_openai_smoke.py scripts/verify_vsf_artifacts.py scripts/benchmark_large_dataset.py
npm run test:e2e:dashboard
make demo-full

Expected optional local connector smokes still skip cleanly when
VSF_POSTGRES_TEST_URL or VSF_MYSQL_TEST_URL is not configured.

Compatibility

• Existing CLI commands, web-runner routes, upload mode, local path mode,
  artifact names, artifact URLs, package outputs, and report paths are
  preserved.
• No hosted backend is introduced. Full browser-driven profiling still requires
  vsf-profiler web on 127.0.0.1.
• The dashboard continues to consume generated artifacts only; it does not read
  raw source CSV rows.