Skip to content

DECISIONS.md

Latest commit

 

History

History
29 lines (21 loc) · 1.34 KB

DECISIONS.md

File metadata and controls

29 lines (21 loc) · 1.34 KB
Raw

Decision Log

D-001: Standalone DatasetOps Repo

  • Status: Accepted
  • Decision: Build datasetops-vision-lab as a standalone repository with apps/web and engine/python.
  • Rationale: The project must not be mixed with FlowForger, portfolio dashboard, or any other project.
  • Consequence: GitHub and Vercel deployment are separate.

D-002: Python Engine First

  • Status: Accepted
  • Decision: Make the Python Audit Engine the canonical audit path and report generator.
  • Rationale: OpenCV/Python is better suited for offline computer vision metrics than browser-only scanning.
  • Consequence: The dashboard imports latest-report.json rather than requiring browser scans first.

D-003: Deterministic Recommendations

  • Status: Accepted
  • Decision: Recommendations use deterministic rule-based problem -> evidence -> action entries.
  • Rationale: Users can trace every recommendation to visible dataset evidence.
  • Consequence: No random AI advice or dataset-name-specific suggestions.

D-004: Transparent V1 Limits

  • Status: Accepted
  • Decision: V1 does not perform semantic understanding, near-duplicate detection, training, or classification.
  • Rationale: The product audits dataset readiness risk, not model correctness.
  • Consequence: Limitations are shown in docs, reports, and UI.