Notebook gallery

atomref ships five explanatory Jupyter notebooks. These are the actual committed .ipynb sources rendered directly by MkDocs—there is no generated Markdown copy or parallel notebook tree.

Documentation builds show the committed Markdown, code, mathematics, saved text, and saved PNG plots. They do not execute or rewrite the notebooks. A separate release check runs each temporary copy in its own bounded standard Jupyter process, fails on execution exceptions or lifecycle timeouts, and discards the temporary outputs without comparing them with the committed files. Kernel startup and cell execution have separate timeouts; a stalled cleanup or process exit cannot block the remaining release gate indefinitely.

User workflows

• Quickstart introduces direct scalar values, provenance-carrying lookup, and dataset discovery.
• Policies and assessment explains ordered restoration, fitted transfers, provenance, and policy summaries.
• Custom sets and discovery builds a user-provided scalar set and inspects the packaged catalog.
• Proatomic density and IAS workflows covers profile provenance, unit-aware evaluation, both pairwise modes, diagnostic cases, formulas, and saved plots.

Supporting numerical study

• IAS method-selection study records the numerical evidence behind the stable boundary and practical minimum contracts. It is supporting information, not an alternative runtime implementation.

The durable scientific summary is also available in Pairwise boundary and IAS-proxy method selection.

Notebook sources

The canonical file for every rendered page is available here for viewing or download:

• 01-quickstart.ipynb
• 02-policies-and-assessment.ipynb
• 03-custom-sets-and-discovery.ipynb
• 04-ias-method-selection-study.ipynb
• 05-proatomic-density-and-ias.ipynb

Run notebooks locally

Install all direct notebook dependencies with:

pip install "atomref[notebooks]"

The notebooks are designed for a standard Python Jupyter kernel. Saved outputs are examples, not a byte-for-byte reproducibility promise: timing, rendering metadata, and PNG bytes may vary when a notebook is deliberately rerun.