RC1 comprehensive validation triage

This table classifies the apparent failures recorded by the external cubedynamics==0.1.0rc1 black-box validation. It separates defects from scientifically useful rejection and from problems in the validation design. The classification was written before this maintainer pass changed runtime code.

Finding	Classification	Maintainer determination
A. A condition followed by mean retains condition/Boolean metadata and generically averages magnitude and threshold	REAL DEFECT	A condition mean is an occurrence proportion summary. Generic averaging of the auxiliary fields has no declared scientific meaning. The canonical reduction should return a summary Dataset whose state is the reduced proportion; condition-definition metadata remains inspectable.
B. Variance retains source units such as degC	REAL DEFECT	Variance units must be squared in both xarray attrs and semantic state. Missing and explicitly unknown units must remain missing/unknown rather than being fabricated.
C1. v.plot() rejects the canonical condition Dataset	REAL DEFECT	Grammar and implementation disagree. A condition Dataset has a canonical state variable and should render it by default.
C2. A condition-frequency Dataset cannot be plotted naturally	REAL DEFECT	This is the combined consequence of A and C1. The state proportion should remain a canonical, directly renderable field.
C3. A 2-D continuous summary constructs a CubePlot but fails in _repr_html_()	REAL DEFECT	Dimensional support must be validated and selected before returning a viewer. A spatial summary should use a real 2-D map renderer; a temporal vector should use a line renderer.
D. Lexcube is absent from a core installation	OPTIONAL DEPENDENCY / ENVIRONMENT	Lexcube correctly remains outside core dependencies. CubeDynamics must translate the raw import failure into installation guidance for the viz extra.
E. Conditions have incompatible DataArray/Dataset representations	REAL DEFECT	Canonical conditions are Datasets exposing state. Threshold conditions may also expose meaningful magnitude and threshold; Boolean overlap has no meaningful magnitude or threshold and should expose only state plus operand/alignment metadata.
F1. Overlap rejects shifted spatial or temporal coordinates	CORRECT SEMANTIC GUARDRAIL	Exact alignment is a scientific invariant and must remain enforced.
F2. A shifted-time input is reported as a dimension-order mismatch	REAL DEFECT	Dataset dimension-map order is not reliable evidence of selected-variable order. Runtime diagnostics must identify temporal coordinates, spatial coordinates, or dimension names accurately; harmless axis order can be normalized.
G. Many unconstrained random verb strings fail	TEST DESIGN ISSUE	Exception count is not a grammar quality metric. Registry-derived legal paths should succeed and known illegal transitions should be rejected with semantic guidance.
H. mean(...) followed by threshold_state(...) was rejected	REAL DEFECT	The manuscript-facing order lesson already teaches both orders as distinct scientific questions. The broader grammar is therefore the repository-consistent design: summaries may be thresholded, producing a new condition and an ORDER_CHANGES_MEANING note.
I. Fire environmental attachment receives a climate cube outside the fire dates	EXPECTED DATA / TEMPORAL MISMATCH	No-overlap rejection is correct. The error should report both ranges, and an overlapping multi-variable regression should prove the positive path.
J. The release routes a documented cube-first FireEventDaily call to a legacy error	REAL DEFECT	The current checkout already binds v.fire_plot to the canonical cube-first implementation, unlike the tested release. Preserve compatibility for fired_daily/event_id, but document FireEventDaily -> FireHull -> attach_environment -> plot as the object path and v.fire_plot(cube, fired_event=...) as the high-level convenience path.

Preserved validation baselines

The successful advertised live noun/source matrix, missing-value handling, Dask laziness, unwrap/repipe interoperability, event detection, flattening, and custom apply behavior are regression constraints. They are not failures and are not reclassified to make this table look more favorable.

Canonical decisions for this pass

• A condition is an xarray Dataset with a Boolean state variable. Threshold and quantile conditions additionally carry scientifically defined magnitude and threshold variables.
• condition | mean(...) returns a summary Dataset containing the reduced floating-point state proportion. It does not implicitly average auxiliary condition fields. Users may select and reduce magnitude explicitly when that is their stated scientific question.
• overlap returns a condition Dataset containing only Boolean state, with exact-alignment and left/right operand metadata.
• mean | threshold_state remains supported because current educational material treats it as a valid, order-sensitive scientific expression.
• General v.plot dispatches by selected DataArray shape: an interactive cube for 3-D time-space data, a static spatial map for 2-D data, and a line plot for a 1-D temporal field. Other shapes fail before a viewer is returned.
• Fire geometry, climate attribution, and rendering remain separate: FireEventDaily.to_hull(), FireHull.attach_environment(...), and FireHull.plot(...) are the preferred object path; v.fire_plot is the cube-first high-level convenience path. The older fire-first keywords remain compatibility inputs, not the preferred teaching route.

Implemented resolution

• Result attrs, semantic state, trace, and explain() now agree for reductions, anomalies, z-scores, conditions, and overlaps. Condition means expose only a floating state proportion; variance units are squared deterministically.
• v.plot() selects canonical condition/event fields and dispatches 3-D cubes, 2-D spatial maps, and 1-D temporal lines before returning a notebook object.
• Exact overlap alignment remains mandatory, with separate temporal, spatial, dimension-name, and harmless-axis-order behavior.
• Missing Lexcube raises install/restart guidance for cubedynamics[viz].
• FireHull.plot() can inspect geometry colored by event day without inventing climate. Environmental colors require an attached or supplied overlapping cube, and mismatches report both time ranges.
• v.fire_plot(cube, fired_event=event) is regression-tested as the public high-level route; compatibility inputs remain available.

The public 0.1.0rc1 artifact predates these fixes. Release-management must assign and review the next candidate version before publication; this triage does not overwrite, upload, tag, or republish rc1.

2026-09-01 naive-user first-use follow-up

An independent clean-environment run found three additional P0 blockers that the broad semantic suite did not exercise:

Finding	Root cause	Resolution and acceptance boundary
Core wheel import failed on Debian 13/aarch64 with libexpat.so.1 missing	cubedynamics.data.sentinel2 eagerly imported Cubo; Cubo imported Rasterio; Rasterio 1.4.4's aarch64 wheel loaded bundled GDAL that retained an unresolved libexpat.so.1 dependency.	Sentinel/Cubo loading is deferred until the Sentinel source is called. A python:3.11-slim x86_64/aarch64 installed-wheel job blocks Cubo/Rasterio/Rioxarray and proves the core public import and first-use path remain usable. This does not certify Sentinel-2 on a system with a broken Rasterio runtime.
Ordinary NetCDF export rejected is_synthetic=False	h5netcdf rejects Boolean attributes, while provenance and noun normalization emitted Python/NumPy Boolean metadata. NetCDF also has no native Boolean variable type for semantic state.	Public source metadata uses portable scalar encodings. The explicit output verb applies canonical JSON/scalar attribute encoding and writes Boolean variables as flagged int8 on a shallow copy. Unsupported opaque objects fail clearly; metadata is never silently dropped.
The first-discovered PRISM example selected monthly semantics	The modern public loader inherited the legacy positional default time_res="ME", while the maintained real NcSS path supports daily requests only; duplicated old getting-started pages omitted freq.	Modern keyword calls default to daily and reject other frequencies before provider access. Current examples state freq="D", old first-use URLs are labeled compatibility bridges, and a bounded live script executes the maintained PRISM request in online CI.

The installed-wheel gate now covers import, noun/source discovery, composition, semantic explanation/validation/trace, dimensional plotting, unwrap, direct continuous export, and condition-safe export. The checksum-pinned external quickstart independently supplies real PRISM retrieval; the online PRISM check is availability evidence, not scientific certification.