Repository language and gallery standards

These requirements supplement AGENTS.md. Read the language sections for Python changes and the gallery section for documentation examples.

Python conventions

• Type hints on all public APIs

• Use list[T], dict[K, V], X | Y — never typing.List, Optional

• src/ layout; explicit __all__ on public packages

• Protocols (PEP 544) over ABCs for structural typing

• @dataclass(frozen=True, slots=True) for value objects

• pathlib for paths; os for env vars only

• Prefer functions over single-use classes

• Fail fast: custom exceptions with contextual messages; zero silent failures

• logging.getLogger(__name__) — never print() or logging.basicConfig()

• Explicit imports and predictable control flow over cleverness; no package-level magic re-exports

• Keep modules small and names descriptive.

Docstrings

Use Google-style docstrings for every public API via sphinx.ext.napoleon. Types belong in signatures only and should not be repeated in docstrings.

Required public API sections:

• Summary

• Description

• Args

• Returns

• Raises

• Example

Documentation gallery

• Keep documentation examples focused on the GenomeSpy visualization API, not statistical analysis or general-purpose data wrangling. Prepare derived example datasets, including p-values, adjusted p-values, transformed values, classifications, and plotting domains, in src/genome_spy/datasets/ or in curated packaged data files. Examples should load the prepared table and demonstrate encodings, marks, transforms, and composition directly.

• Prefer Altair-style composition operators in examples when they express the intended structure clearly: + for layers, & for vertical concatenation, and | for horizontal concatenation.

• Documentation gallery embeds must use the direct static GenomeSpy bundle import followed by embed(c, spec, { bare: true }). Keep the chart container direct and avoid transform or resize wrappers around the embedded canvas; this is the known-working path for all gallery plots, including sashimi.

• Keep adapted gallery pages self-contained. Link the corresponding official GenomeSpy example as further reading, not as a substitute for local explanation.

• Put extended prose in an optional same-stem Markdown companion under docs/examples/.

• For the same data and processing, reuse the official GenomeSpy example’s concise provenance or disclaimer wording. Update it when the sample, region, subset, preprocessing, or intended use differs.

• Keep provenance as brief as possible. Include only details needed to identify the data, disclose material processing, satisfy attribution or licensing, or prevent misuse. Use a note-styled admonition titled Data use and provenance; do not repeat it as a section heading.

• Adapt What to notice and implementation prose to the chart actually rendered. Verify every claim about data, encodings, transforms, interaction, and composition against the checked-in Python source.

• Explicitly distinguish processing performed in Python from GenomeSpy’s browser-side dataflow. State whether an example loads a prepared packaged table, derives or filters data in Python, or sends data to GenomeSpy for declarative transforms.

• Do not describe packaged derived data as raw source data or imply that .transform_*() methods execute in Python. The Python API authors and serializes those transform definitions; GenomeSpy executes them while rendering or interacting with the visualization in the browser.