Make automatic_dependency_source_indexing_disabled the benchmark default and add the paired enabled profile. Scope candidate_native_configuration to older candidates, reject unbacked capability claims, and pin all latest-product graph capabilities so one-factor ablations remain comparable.
Remove inherited CBM_* product variables, retain isolated per-case caches, record native worker selection, and emit requested/effective configuration facts without inventing defaults for candidate-native or legacy reports. Keep retained historical reports readable.
Verification: uv run --with pytest python -m pytest tests/test_benchmark_incremental_speed.py tests/test_benchmark_experiments.py tests/test_benchmark_experiments_shim.py tests/test_autotune.py tests/test_summarize_benchmark_results.py (184 passed, 1 skipped); ruff format --check and ruff check passed for all touched Python files.
Signed-off-by: Andrew Hundt <ATHundt@gmail.com>
Andrew Hundt
8d415b9fd2