[imatiach-msft]

Summary

Closes #108.
Closes #91.

• Add introspection-based API contract tests for nine explainers and seven explanation classes, pinning parameter order and separating public methods, properties, and other class attributes.
• Validate explicit feature-name counts at initialization and evaluation boundaries, against the original input before transformation or subsetting.
• Preserve inferred names, index handling, reused dataset wrappers, legacy serialization, and supported SHAP statistical/masker backgrounds. Read width metadata without materializing DataFrames unnecessarily.

Validation

python tests\run_tests.py --compatibility-only -q:

Environment	Result
Python 3.14 / SHAP 0.54.1	198 passed
Python 3.11 / SHAP 0.48.0 / NumPy 2.0.2 with CI minimum dependency constraints	192 passed, 6 skipped
Python 3.10 / SHAP 0.46.0	192 passed, 6 skipped

The six older-SHAP skips cover bare statistical dictionaries that upstream SHAP does not support. Tests probe upstream support; interpret-community failures are not suppressed.

• 29 existing index, datetime, summarized-background, subset, and transformation integration tests passed.
• Changed-file isort and flake8 checks, plus git diff --check, passed.
• Mutation checks verified that feature-validation and API-contract regressions fail the new tests.
• Successful Deep/GPU integrations were not run locally; malformed inputs are validated before backend initialization.
• All 16 GitHub checks passed on the single signed-off commit 69202f1: full Python 3.10-3.14 tests, all four NumPy compatibility jobs, dashboard, lint, CodeQL/code scan, DCO, and both Codecov checks. The initial partial-coverage failures cleared after the unit-test coverage upload.

Recursive review

Eight independent full-diff rounds used GPT-6.1 Sol, Claude Sonnet 5.5, and Grok 4.7, each explicitly requesting maximum reasoning. Runtime-reported models matched; effective reasoning effort was not separately reported. All three final reviewers reported no actionable findings.

One documented concern was rejected: preserving incidental acceptance of None/list initialization backgrounds with explicit names is outside TabularExplainer's documented ndarray/DataFrame/csr input contract.

Final review status: clean with documented rejected findings.

[codecov]

Codecov Report

❌ Patch coverage is 98.11321% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 88.34%. Comparing base (1c9d179) to head (69202f1).

Files with missing lines	Patch %	Lines
python/interpret_community/dataset/decorator.py	97.67%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #609      +/-   ##
==========================================
+ Coverage   88.07%   88.34%   +0.26%     
==========================================
  Files          59       59              
  Lines        4077     4127      +50     
==========================================
+ Hits         3591     3646      +55     
+ Misses        486      481       -5     

Flag	Coverage Δ
dashboard	42.64% <71.69%> (+0.33%)	⬆️
unittests	87.61% <98.11%> (+0.27%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.