Add kriging diagnostic outputs to the task SDK client

[amirabbas-jalali-bentley]

Description

Adds support for the kriging task's diagnostic outputs to the evo-compute SDK client.

The kriging task can write per-location diagnostics onto the target object alongside the estimate — kriging variance, slope of regression, sample counts, search distances, and so on. These live under target.diagnostics in the task schema, and were previously not reachable through the SDK.

Parameters

KrigingParameters gains a diagnostics field taking a KrigingDiagnostics. It is folded into target.diagnostics by the existing model serializer, the same way source_filter and target_filter are already folded into source.filter / target.filter — so callers keep passing a typed attribute (or a Target) for target and do not have to hand-build the nested object.

Every diagnostic accepts:

• True — create an attribute using the recommended name (RECOMMENDED_DIAGNOSTIC_NAMES, e.g. KV, SoR, NS)
• a string — create an attribute with that name
• a typed attribute from the target object — update it if it exists, otherwise create it
• an explicit CreateAttribute / UpdateAttribute

params = KrigingParameters(
    source=pointset.attributes["CU_pct"],
    target=block_model.attributes["CU_estimate"],
    variogram=variogram,
    search=SearchNeighborhood(ellipsoid=search_ellipsoid, max_samples=16),
    diagnostics=KrigingDiagnostics(
        kriging_variance=True,        # -> "KV"
        slope_of_regression=True,     # -> "SoR"
        num_samples="sample_count",   # -> "sample_count"
    ),
)

The recommended names match the conventions used elsewhere in the product so results are familiar once imported. They are the schema's suggested names rather than server-side defaults, so the SDK always sends an explicit attribute name.

Validation

Two rules the task enforces server-side are checked before the payload is built, so mistakes surface as a ValidationError on KrigingParameters rather than a failed job:

• Every output needs its own attribute. Duplicate create names or update references are rejected, including a diagnostic colliding with target.attribute. Create names and update references are compared separately, since they are distinct namespaces.
• Diagnostics belong to the target object. A typed attribute is reduced to a name or reference, which is then resolved against target.object — so an attribute taken from a different object would silently resolve to nothing, or to the wrong attribute. The originating object is now remembered during validation and compared with the target, ignoring any ?version= suffix. Attributes that carry no parent object cannot be checked and are left alone.

Results

The task returns every diagnostic key, with null for the ones that were not requested. New KrigingDiagnosticsResult and KrigingTargetResult models parse that, and KrigingResult.diagnostics exposes only the diagnostics that were actually written, keyed by diagnostic name:

for diagnostic, attribute in result.diagnostics.items():
    print(f"{diagnostic}: {attribute.name}")

They are also listed in the result summary. Because diagnostics are written onto the target object, result.to_dataframe() returns them alongside the estimate.

Other changes

• attribute_spec() factored out of _validate_target_attribute in tasks/common/source_target.py and shared, so a diagnostic accepts the same typed attributes as a target does.
• KrigingDiagnostics exported from evo.compute.tasks and evo.compute.tasks.geostatistics.
• tests/test_tasks.py fixtures build a KrigingTargetResult instead of a bare TaskTarget, following the retyped KrigingResultModel.target.

Backwards compatibility

No breaking change to the SDK's public surface — diagnostics is optional and defaults to None, so existing kriging calls produce an identical payload.

Testing

• uv run --package evo-compute pytest packages/evo-compute/tests — 440 passed (27 new tests covering the accepted shorthands, the serialized payload, the parsed result, and the two validation rules).
• ruff check and ruff format --check clean across the repository.
• The generated payload and a realistic result payload were validated against the kriging task's JSON schema with jsonschema (Draft 2020-12) — no errors on either side.
• Notebooks were verified to parse; they need live Evo credentials to execute, so they were not run end to end.

Documentation

• packages/evo-compute/docs/examples/kriging.ipynb — new "Request Diagnostics" section.
• code-samples/geoscience-objects/running-kriging-compute/ — the walkthrough now requests diagnostics, prints the attributes that were written, and includes them in the results DataFrame; the markdown documents the shorthands and lists all 16 diagnostics with their recommended names.

The generated API reference under mkdocs/ is left to the documentation workflow.

Checklist

• I have read the contributing guide and the code of conduct

[Copilot]

🟡 Changes recommended

Output collisions and cross-object attributes are not validated, and generated documentation links target missing pages.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds typed kriging diagnostic outputs to the compute SDK, including request serialization, result parsing, tests, and examples.

Changes:

• Adds diagnostic selectors, recommended names, and target serialization.
• Exposes returned diagnostic attributes through kriging results.
• Updates exports, tests, notebooks, generated documentation, and dependencies.

File summaries

File	Description
uv.lock	Updates evo-blockmodels lock version.
packages/evo-compute/tests/test_tasks.py	Uses the specialized kriging target result.
packages/evo-compute/tests/geostatistics/test_kriging_tasks.py	Tests diagnostic inputs, serialization, and results.
packages/evo-compute/src/evo/compute/tasks/geostatistics/kriging.py	Implements diagnostic request and result models.
packages/evo-compute/src/evo/compute/tasks/geostatistics/__init__.py	Exports KrigingDiagnostics.
packages/evo-compute/src/evo/compute/tasks/common/source_target.py	Extracts typed attribute conversion.
packages/evo-compute/src/evo/compute/tasks/common/__init__.py	Exports attribute_spec.
packages/evo-compute/src/evo/compute/tasks/__init__.py	Adds the top-level diagnostics export.
packages/evo-compute/docs/examples/kriging.ipynb	Demonstrates requesting diagnostics.
mkdocs/site/packages/evo-files/index.html	Updates generated previous-page navigation.
mkdocs/site/packages/evo-compute/typed-objects/source-target/UpdateAttribute.html	Updates generated next-page navigation.
mkdocs/site/packages/evo-compute/typed-objects/kriging/SimpleKriging.html	Updates generated navigation.
mkdocs/site/packages/evo-compute/typed-objects/kriging/OrdinaryKriging.html	Updates generated navigation.
mkdocs/site/packages/evo-compute/typed-objects/kriging/KrigingRunner.html	Updates generated navigation.
mkdocs/site/packages/evo-compute/typed-objects/kriging/KrigingResultModel.html	Documents the specialized target result.
mkdocs/site/packages/evo-compute/typed-objects/kriging/KrigingResult.html	Documents result diagnostics.
mkdocs/site/packages/evo-compute/typed-objects/kriging/KrigingParameters.html	Documents diagnostic parameters.
mkdocs/site/packages/evo-compute/typed-objects/kriging/KrigingMethod.html	Updates generated navigation.
code-samples/geoscience-objects/running-kriging-compute/running-kriging-compute.ipynb	Extends the walkthrough with diagnostics.
code-samples/geoscience-objects/running-kriging-compute/README.md	Lists diagnostics in the walkthrough scope.

Review details

Suppressed comments (1)

mkdocs/site/packages/evo-compute/typed-objects/kriging/OrdinaryKriging.html:42

• RECOMMENDED_DIAGNOSTIC_NAMES.html is absent from the checked-in kriging API directory, making this Next link a 404. Regenerate and commit the missing API page.

                                <a rel="next" href="RECOMMENDED_DIAGNOSTIC_NAMES.html" class="nav-link">

• Files reviewed: 19/20 changed files
• Comments generated: 8
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.