Architecture

System-level understanding of Datasculpt.

Design Goals

1. Determinism — Same input → same output
2. Evidence-based — Every inference is justified
3. Focused — Structure only, not semantics
4. Minimal dependencies — Core requires only pandas
5. Auditable — Complete decision trails

Non-Goals

Datasculpt explicitly does not:

• Replace data catalogs — It produces metadata for catalogs, not stores it
• Perform ETL — It analyzes structure, not transforms data
• Define semantics — That's Invariant's job
• Use LLMs for decisions — Determinism requires rule-based inference
• Handle data quality — It reports issues, not fixes them

Architecture Guides

Guide	Focus
Pipeline	Inference stages and data flow
Design Principles	Core principles and rationale
Scope Boundaries	What's in and out of scope