Change review checklist

Use the relevant parts of this checklist before creating a checkpoint, opening a pull request, or handing work to another contributor.

Scientific contract

[ ] The physical quantity and approximation are stated.
[ ] Units, normalization, tensor convention, axis order, and dtype are explicit.
[ ] Current behaviour was characterized before a numerical refactor.
[ ] Analytical, legacy, published, experimental, or independent references
were identified where required.
[ ] Tolerances have rationale and boundary tests.
[ ] Invalid, unresolved, not-applicable, and extrapolated states are distinct.
[ ] Uncertainty and covariance semantics are documented.
[ ] No scientific data are rounded for storage or internal use.

Architecture

[ ] Core code has no frontend or workflow imports.
[ ] Scientific modules do not import one another.
[ ] Active objects are classes; passive contracts are dataclasses.
[ ] Long numerical loops use simple callbacks rather than Quantas events.
[ ] Module plot builders do not import Matplotlib.
[ ] The CLI calls quantas.api.
[ ] Optional plotting dependencies do not leak into base calculations.

Inputs and interfaces

[ ] File recognition and completion checks are separate.
[ ] External units and conventions are converted explicitly.
[ ] Source provenance and relevant external-code settings are preserved.
[ ] Complete, incomplete, malformed, and variant fixtures are tested.
[ ] Normalized inputs have documented shapes and units.

Results and persistence

[ ] The typed result contains full-precision values and aligned masks.
[ ] New HDF5 fields have unit and description metadata.
[ ] Writer and reader were updated together.
[ ] Current and supported historical schemas are explicit.
[ ] Write/read round-trip tests pass.
[ ] Missing values are not confused with zero.
[ ] Metadata-based dispatch still works.

Events and errors

[ ] Meaningful stages, warnings, and failures emit appropriate levels.
[ ] Progress events are not persisted.
[ ] Event data are lightweight and serializable.
[ ] Errors are actionable and not hidden by broad exception handling.
[ ] A warning is not used where the result should be rejected.

Reports, plots, and exports

[ ] Report tables contain raw values, units, and deterministic ordering.
[ ] Plain reports contain no ANSI or live-progress artifacts.
[ ] Plot specifications are frontend-neutral.
[ ] Plot resolution and renderer quality controls are distinguished from
scientific convergence controls.
[ ] Exports preserve documented precision, units, and provenance.

Public API and CLI

[ ] Public names are reviewed and added explicitly to __all__.
[ ] Every public symbol has a technical docstring and one API-reference entry.
[ ] Registry capabilities and types are current.
[ ] CLI help explains each new option.
[ ] API and CLI execute the same numerical path.
[ ] Output paths, overwrite behaviour, and exit status are tested.

Tests

[ ] Focused unit/module tests pass.
[ ] Architecture and public-surface tests pass.
[ ] HDF5 and export tests pass.
[ ] CLI/API equivalence is covered.
[ ] Plot-specification and renderer tests are separated.
[ ] A real-data or scientific regression exists when appropriate.
[ ] Ruff, Mypy, and compileall pass.
[ ] The complete staged test runner passes.

Documentation and citations

[ ] Theory, workflow, tutorial, format, CLI, and API pages were inspected.
[ ] Scientific references are in the canonical registry.
[ ] Bibliography and tutorial assets were regenerated when affected.
[ ] RST links, downloads, images, and title underlines are valid.
[ ] Sphinx builds with warnings as errors.
[ ] Commands and Python examples were executed against the snapshot.

Packaging

[ ] New files are included in the intended distribution artifact.
[ ] Wheel and source distribution build successfully.
[ ] Twine validation passes.
[ ] Clean-install distribution checks pass.
[ ] py.typed and the console entry point remain available.
[ ] Platform-specific paths and subprocesses are portable.

A checkbox is not evidence by itself. Record the command, test, dataset, or reference that demonstrates each important scientific and release claim.