test_cache_claims_require_hits_and_clean_control (test_cache.CacheTests.test_cache_claims_require_hits_and_clean_control) ... ok
test_cases_balanced_unique_and_frozen_targets (test_cache.CacheTests.test_cases_balanced_unique_and_frozen_targets) ... ok
test_luna_arms_only_differ_in_breakpoint (test_cache.CacheTests.test_luna_arms_only_differ_in_breakpoint) ... ok
test_missing_cost_and_cache_counters_are_unknown (test_cache.CacheTests.test_missing_cost_and_cache_counters_are_unknown) ... ok
test_mock_run_separates_prime_and_respects_missing_cost (test_cache.CacheTests.test_mock_run_separates_prime_and_respects_missing_cost) ... ok
test_policy_boundaries_and_exceptions (test_cache.CacheTests.test_policy_boundaries_and_exceptions) ... ok
test_prefix_fixed_suffix_varies (test_cache.CacheTests.test_prefix_fixed_suffix_varies) ... ok
test_boolq_labels_and_rebuild_against_full_source_when_available (test_datasets.FrozenDatasetTests.test_boolq_labels_and_rebuild_against_full_source_when_available) ... ok
test_boolq_sampling_and_attribution (test_datasets.FrozenDatasetTests.test_boolq_sampling_and_attribution) ... ok
test_invalid_schema_is_rejected (test_datasets.FrozenDatasetTests.test_invalid_schema_is_rejected) ... ok
test_manifest_hash_and_counts (test_datasets.FrozenDatasetTests.test_manifest_hash_and_counts) ... ok
test_model_inputs_have_no_evaluator_metadata (test_datasets.FrozenDatasetTests.test_model_inputs_have_no_evaluator_metadata) ... ok
test_policy_distractors_cannot_change_truth (test_datasets.FrozenDatasetTests.test_policy_distractors_cannot_change_truth) ... ok
test_policy_truth_independently (test_datasets.FrozenDatasetTests.test_policy_truth_independently) ... ok
test_probability_truth_by_outcome_enumeration (test_datasets.FrozenDatasetTests.test_probability_truth_by_outcome_enumeration) ... ok
test_synthetic_reproducibility (test_datasets.FrozenDatasetTests.test_synthetic_reproducibility) ... ok
test_broken_jsonl_reports_line (test_report.ReportTests.test_broken_jsonl_reports_line) ... ok
test_empty_and_all_failed_groups_are_reportable (test_report.ReportTests.test_empty_and_all_failed_groups_are_reportable) ... ok
test_exact_probability_scores_use_expected_outcomes (test_report.ReportTests.test_exact_probability_scores_use_expected_outcomes) ... ok
test_generate_files_escape_untrusted_text_and_accept_string_protocol (test_report.ReportTests.test_generate_files_escape_untrusted_text_and_accept_string_protocol) ... ok
test_labeled_metrics_and_invalid_denominator (test_report.ReportTests.test_labeled_metrics_and_invalid_denominator) ... ok
test_missing_cost_not_zero_and_failed_cost_included (test_report.ReportTests.test_missing_cost_not_zero_and_failed_cost_included) ... ok
test_mixed_target_kinds_rejected (test_report.ReportTests.test_mixed_target_kinds_rejected) ... ok
test_nonfinite_and_out_of_range_predictions_invalid (test_report.ReportTests.test_nonfinite_and_out_of_range_predictions_invalid) ... ok
test_paired_bootstrap_direction_and_repeat_clustering (test_report.ReportTests.test_paired_bootstrap_direction_and_repeat_clustering) ... ok
test_paired_excludes_failures_and_conflicting_targets (test_report.ReportTests.test_paired_excludes_failures_and_conflicting_targets) ... ok
test_warmups_excluded_from_quality_but_not_total_billing (test_report.ReportTests.test_warmups_excluded_from_quality_but_not_total_billing) ... ok
test_cache_arm_retains_model_identity (test_runner.RunnerTests.test_cache_arm_retains_model_identity) ... ok
test_exception_cannot_echo_key (test_runner.RunnerTests.test_exception_cannot_echo_key) ... ok
test_labels_do_not_reach_models (test_runner.RunnerTests.test_labels_do_not_reach_models) ... ok
test_missing_cost_is_unknown_not_zero (test_runner.RunnerTests.test_missing_cost_is_unknown_not_zero) ... ok
test_other_models_rejected (test_runner.RunnerTests.test_other_models_rejected) ... ok
test_strict_invalid_values (test_runner.RunnerTests.test_strict_invalid_values) ... ok
test_truncated_output_is_failure_but_charged (test_runner.RunnerTests.test_truncated_output_is_failure_but_charged) ... ok

----------------------------------------------------------------------
Ran 34 tests in 0.066s

OK