Notebook Analytics Workbench

## What this collection sets up

This collection is a practical notebook analytics workbench: write analysis in
reviewable notebooks, query local files without standing up a warehouse,
transform tabular data efficiently, load approved datasets, validate important
columns, and share reviewed results as lightweight apps.

It is not a privacy-first research workflow and it is not a full data
engineering platform. The focus is the analyst loop from raw files or approved
datasets to reproducible notebook, validated result, and shareable report.

## Layers

### 1. Notebook and local query layer

- **marimo** keeps Python notebooks in a git-friendly source format and supports
  notebook, script, app, SQL, and reactive execution workflows.
- **duckdb** gives notebooks and scripts an embedded analytical SQL engine for
  local files, Parquet, CSV, joins, and single-file analysis.
- **polars** provides a fast DataFrame engine for filtering, joins,
  transformations, lazy execution, streaming, and export preparation.

### 2. Dataset intake and quality checks

- **hugging-face-datasets** loads, streams, inspects, and preprocesses Hub or
  local datasets with cache and split/configuration controls.
- **great-expectations** records data-quality expectations, validation
  definitions, checkpoints, and Data Docs before outputs are trusted.

### 3. Sharing reviewed analysis

- **streamlit** turns reviewed scripts into interactive data apps, dashboards,
  reports, and analytical interfaces when a notebook needs a human-facing view.

## Suggested order

Start with Marimo so analysis lives in reviewable source. Add DuckDB for local
SQL against files and Polars for DataFrame transformations. Load external or
large datasets only after their source, license, schema, and cache behavior are
approved. Add Great Expectations checks around key columns and metrics before
sharing results. Use Streamlit last, after the analysis is stable and the data
exposure rules are clear.

## Analysis checklist

- [ ] {"task": "Sources are inventoried", "description": "Files, datasets, licenses, revisions, credentials, and cache paths are known"}
- [ ] {"task": "Notebook is reproducible", "description": "Dependencies, paths, parameters, and execution order are clear enough for review"}
- [ ] {"task": "Queries are bounded", "description": "Large scans, joins, exports, and remote downloads have limits and expected sizes"}
- [ ] {"task": "Data quality is checked", "description": "Important columns, row counts, nulls, ranges, and joins have validation expectations"}
- [ ] {"task": "Outputs are reviewed", "description": "Charts, CSVs, notebooks, Data Docs, and apps are checked before sharing"}
- [ ] {"task": "Retention is explicit", "description": "Caches, DuckDB files, Streamlit state, screenshots, and exports have cleanup rules"}

## Source and references

- Marimo getting started: https://docs.marimo.io/getting_started/
- DuckDB clients overview: https://duckdb.org/docs/stable/clients/overview
- Polars getting started: https://docs.pola.rs/user-guide/getting-started/
- Hugging Face Datasets documentation: https://huggingface.co/docs/datasets/index
- Great Expectations introduction: https://docs.greatexpectations.io/docs/core/introduction/
- Streamlit get started: https://docs.streamlit.io/get-started

## Duplicate check

Checked existing collections, tools, skills, commands, hooks, guides, open PRs,
closed PRs, and issue history for `notebook-analytics-workbench`, data analysis
notebook workflow, Marimo, DuckDB, Polars, Hugging Face Datasets, Great
Expectations, Streamlit, analytics workbench, and notebook dashboards.
`privacy-first-research-workflow` uses Marimo, DuckDB, and Polars inside a
privacy/redaction/research workflow. `data-engineering-suite` focuses on ETL,
databases, cloud services, and pipeline infrastructure. This entry is narrower:
it covers the analyst notebook loop from data intake to local query,
transformation, quality checks, and reviewed sharing.

## Disclosure

Editorial collection. No paid placement or affiliate link is used.