90 difficulty-balanced tasks

Two tasks per difficulty level, five levels per benchmark: 10 tasks from each of nine scored benchmarks. Pairs favor different intents and technical domains. Difficulty levels are relative to each benchmark. No scored trial data: original SWE-bench, Pro V2, SWE Atlas, RealSWE; these four are omitted. Raw tags remain unchanged.

#	Benchmark	Task intent	Technical domain	Technology stack	Context requirements	Change scope	Reasoning demands	Specification completeness	Constraints	Execution and verification	Difficulty (within benchmark)
1	deep	feature_implementation	static_analysis	Python	repository_exploration	public_interface_schema_change	algorithmic_reasoning; state_concurrency_reasoning	expected_behavior_stated	unknown	reproduction_check; edge_case_testing	very hard; mean reward 0.0550; 200 scored trials
2	deep	feature_implementation	compilers_language_tooling	CSS	repository_exploration	public_interface_schema_change	algorithmic_reasoning	expected_behavior_stated	unknown	reproduction_check; edge_case_testing	very hard; mean reward 0.2750; 200 scored trials
3	deep	feature_implementation	backend_api	unknown	repository_exploration	unknown	algorithmic_reasoning; state_concurrency_reasoning	expected_behavior_stated	unknown	reproduction_check; edge_case_testing	hard; mean reward 0.3250; 200 scored trials
4	deep	feature_implementation	terminal_ui	JavaScript	repository_exploration	unknown	algorithmic_reasoning; state_concurrency_reasoning	expected_behavior_stated; explicit_requirements	compatibility	reproduction_check; edge_case_testing	hard; mean reward 0.3400; 200 scored trials
5	deep	feature_implementation	data_serialization	Python	repository_exploration	public_interface_schema_change	algorithmic_reasoning; state_concurrency_reasoning	expected_behavior_stated	unknown	reproduction_check; edge_case_testing	medium; mean reward 0.6000; 200 scored trials
6	deep	feature_implementation	cli_tooling	unknown	repository_exploration	unknown	algorithmic_reasoning; state_concurrency_reasoning	expected_behavior_stated; environment_supplied	compatibility	reproduction_check; edge_case_testing	medium; mean reward 0.5250; 200 scored trials
7	deep	feature_implementation	cloud_infrastructure	Go	repository_exploration	public_interface_schema_change	algorithmic_reasoning; state_concurrency_reasoning	expected_behavior_stated	performance	reproduction_check; edge_case_testing	easy; mean reward 0.6850; 200 scored trials
8	deep	feature_implementation	data_processing	Go	repository_exploration	public_interface_schema_change	algorithmic_reasoning	expected_behavior_stated	unknown	reproduction_check; edge_case_testing	easy; mean reward 0.6250; 200 scored trials
9	deep	feature_implementation	developer_tools	GitHub_Actions	repository_exploration	unknown	algorithmic_reasoning	expected_behavior_stated	unknown	reproduction_check; edge_case_testing	very easy; mean reward 0.8250; 200 scored trials
10	deep	feature_implementation	databases_storage	SQL	repository_exploration	public_interface_schema_change	algorithmic_reasoning	expected_behavior_stated; explicit_requirements	unknown	artifact_verification; reproduction_check; edge_case_testing	very easy; mean reward 0.8600; 200 scored trials
11	frontier	data_compression	audio_processing; data_compression	Python	unknown	local_edit	algorithmic_reasoning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	compatibility; dependency_restrictions	artifact_verification	very hard; mean reward 0.0230; 50 scored trials
12	frontier	performance_optimization	graphics_image_processing; performance_optimization	Zig; C_ABI; SIMD	external_api_documentation	coordinated_edits_within_component	algorithmic_reasoning; architectural_tradeoffs	explicit_requirements; location_supplied; environment_supplied	performance; compatibility; dependency_restrictions	performance_measurement; integration_tests	very hard; mean reward 0.0681; 50 scored trials
13	frontier	reimplementation	language_tooling; performance_optimization	x86_64_assembly; C_ABI; libc	external_api_documentation	unknown	state_concurrency_reasoning; architectural_tradeoffs	expected_behavior_stated; location_supplied; environment_supplied	performance; compatibility; dependency_restrictions	integration_tests; performance_measurement	hard; mean reward 0.1251; 50 scored trials
14	frontier	model_training	machine_learning; mathematics	Qwen3-14B; PEFT	external_api_documentation	unknown	algorithmic_reasoning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	dependency_restrictions	artifact_generation	hard; mean reward 0.1243; 50 scored trials
15	frontier	model_training	machine_learning; games	Qwen3-8B; PEFT_LoRA	external_api_documentation	local_edit	algorithmic_reasoning	explicit_requirements; expected_behavior_stated; location_supplied; environment_supplied	dependency_restrictions; compatibility	artifact_generation	medium; mean reward 0.1776; 50 scored trials
16	frontier	reimplementation	scientific_software; numerical_computing	Rust	external_api_documentation	unknown	algorithmic_reasoning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	compatibility; dependency_restrictions	model_evaluation	medium; mean reward 0.1768; 50 scored trials
17	frontier	reimplementation	game_development; audio_processing	C; Game_Boy_Advance	external_api_documentation	unknown	state_concurrency_reasoning	expected_behavior_stated; location_supplied; environment_supplied	compatibility; dependency_restrictions	integration_tests	easy; mean reward 0.2891; 50 scored trials
18	frontier	feature_implementation	scientific_software; statistical_inference	unknown	external_api_documentation	unknown	statistical_reasoning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	compatibility; dependency_restrictions	model_evaluation	easy; mean reward 0.3607; 50 scored trials
19	frontier	feature_implementation	quantum_computing	Python	repository_architecture	local_edit	algorithmic_reasoning; architectural_tradeoffs	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	performance; dependency_restrictions	evaluation_harness	very easy; mean reward 0.6096; 50 scored trials
20	frontier	reimplementation	compilers_language_tooling; digital_logic	Swift; Foundation	supplied_snippet	unknown	state_concurrency_reasoning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	dependency_restrictions; compatibility	unit_tests	very easy; mean reward 0.5523; 50 scored trials
21	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	very hard; mean reward 0.9463; 21 scored trials
22	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	very hard; mean reward 0.9658; 21 scored trials
23	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	hard; mean reward 0.9920; 21 scored trials
24	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	hard; mean reward 0.9975; 21 scored trials
25	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	medium; mean reward 0.9978; 21 scored trials
26	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	medium; mean reward 0.9991; 21 scored trials
27	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	easy; mean reward 0.9999; 21 scored trials
28	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	easy; mean reward 0.9996; 21 scored trials
29	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	very easy; mean reward 1.0000; 21 scored trials
30	program	reimplementation; reverse_engineering	unknown	unknown	supplied_binary; bundled_documentation	whole_program_implementation	black_box_behavioral_inference	location_supplied; environment_supplied; behavior_specified_by_executable	compatibility; no_internet; source_retrieval_prohibited; binary_wrapping_prohibited; no_decompilation_or_tracing; dependency_setup_required; fresh_build_required; artifact_exclusion	behavioral_testing; build_checking	very easy; mean reward 1.0000; 21 scored trials
31	swepro1	bug_fix; feature_implementation	web_application; file_management; data_cleanup	NodeBB; Node.js; filesystem	supplied_reproduction; existing_behavior_description	public_interface_change; whole_program_implementation	filesystem_cleanup; path_validation; account_deletion	explicit_requirements; interface_specification	allowed_upload_path_constraint; handle_missing_file_constraint; supported_image_extension_constraint	reproduction_check; file_removal_verification; account_deletion_verification	very hard; mean reward 0.0000; 14 scored trials
32	swepro1	refactoring	text_parsing; expression_language; security	Teleport; Go; AST; regular expressions	existing_codebase	cross_component_change	expression_ast_evaluation_reasoning; namespace_and_variable_validation_reasoning; function_arity_and_type_validation_reasoning	expected_behavior_stated; explicit_requirements; location_supplied	bounded_expression_depth; reject_unknown_or_unsupported_constructs; empty_interpolation_is_not_found	behavioral_testing	very hard; mean reward 0.0000; 11 scored trials
33	swepro1	bug_fix	configuration_management; cloud_integration	Vuls; Go; TOML; UUID	existing_codebase	local_edit	conditional_persistence_reasoning; host_container_identity_link_reasoning	expected_behavior_stated; explicit_requirements	reuse_valid_existing_uuids; write_config_only_when_values_change; container_keeps_host_uuid	behavioral_testing	hard; mean reward 0.0714; 14 scored trials
34	swepro1	feature_implementation	music_library_media_scanning; album_artwork_path_storage; database_schema_migration	Navidrome; Go; filepath	directory_scan_and_album_refresh_context; incremental_rescan_and_directory_deletion_context	whole_program_implementation; public_interface_change	per_directory_image_inventory; path_joining_and_platform_list_separator; directory_map_propagation_through_rescan	explicit_requirements; interface_specification	preserve_other_folder_scan_counters; sorted_deduplicated_media_directories; rescan_media_after_schema_migration	multiple_directory_image_path_join_case_check; incremental_rescan_directory_map_case_check; folder_change_signature_case_check	hard; mean reward 0.0714; 14 scored trials
35	swepro1	bug_fix; feature_implementation; conflicting_requirements	trusted_cluster_heartbeat_status; audit_recording_and_upload	Teleport; Go	RemoteCluster_tunnel_connection_context; legacy_audit_session_recording_context	whole_program_implementation; public_interface_change	preserve_last_heartbeat_after_connection_removal; legacy_recording_unpack_detection; upload_checkpoint_and_completion_management	conflicting_requirements; explicit_requirements; interface_specification	retain_latest_valid_heartbeat_when_offline; only_persist_on_status_or_new_heartbeat_change; legacy_unpacked_recording_skips_download	offline_remote_cluster_heartbeat_case_check; unpacked_legacy_recording_download_skip_case_check	medium; mean reward 0.2308; 13 scored trials
36	swepro1	type_annotation_refactoring; code_cleanup	list_data_model_and_seed_processing; static_type_analysis	Open Library; Python; TypedDict	polymorphic_list_seed_value_context; subject_pseudo_key_normalization_context	whole_program_implementation; public_interface_change	typed_dict_and_return_annotation_design; seed_type_guarding_and_normalization; duplicate_detection_by_normalized_key	explicit_requirements; interface_specification	support_Thing_SeedDict_and_SeedSubjectString_inputs; preserve_export_list_three_named_categories	subject_key_normalization_case_check; all_seed_forms_add_remove_case_check; typed_export_list_shape_case_check	medium; mean reward 0.1429; 14 scored trials
37	swepro1	dependency_injection_refactoring; bug_fix	Subsonic_API_router_construction; playback_service_integration	Navidrome; Go	router_test_instantiation_context; playback_server_dependency_context	public_constructor_interface_change; localized_dependency_wiring_change	dependency_injected_playback_server_construction; generated_wire_injector	explicit_requirements; function_interface_specification	preserve_existing_Subsonic_API_functionality; integrate_playback_service_when_provided	router_instantiates_with_updated_signature_case_check; playback_dependency_integration_case_check	easy; mean reward 0.5385; 13 scored trials
38	swepro1	security_fix	email_attachment_sanitization; cross_site_scripting_prevention	Tutanota; TypeScript; SVG; XML	inline_email_SVG_attachment_processing_context; user_opened_image_must_not_execute_embedded_script_context	whole_program_implementation; public_interface_change	SVG_XML_parsing_and_script_node_removal; safe_content_canonicalization; inline_attachment_pipeline_integration	explicit_requirements; method_interface_specification	non_SVG_attachments_return_unchanged; malformed_XML_returns_empty_data_with_original_metadata; preserve_visual_non_executable_elements_and_attributes	embedded_script_removed_case_check; invalid_UTF8_XML_empty_data_case_check; non_SVG_identity_case_check	easy; mean reward 0.4615; 13 scored trials
39	swepro1	bug_fix	plugin_management; input_validation	NodeBB; Node.js	existing_behavior_description	local_edit	identifier_validation	expected_behavior_stated; explicit_requirements	reject_invalid_identifier_before_mutation	invalid_identifier_case_verification	very easy; mean reward 0.5714; 14 scored trials
40	swepro1	refactoring; security_hardening	automation_tooling; input_parsing	Ansible; Python; JSON	existing_behavior_description; repository_code	whole_program_implementation	safe_data_parsing; deprecation_signaling	explicit_requirements; expected_behavior_stated	no_arbitrary_code_execution; accepted_input_forms_preserved	valid_input_form_verification; unsafe_expression_rejection_check; deprecation_warning_check	very easy; mean reward 0.9167; 12 scored trials
41	terminal21	semantic_search	information_retrieval; text_embeddings	MTEB; BGE	supplied_documents	result_file	similarity_ranking	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	model_revision; package_version	model_loading; result_artifact	very hard; mean reward 0.5545; 110 scored trials
42	terminal21	protein_design	protein_engineering; molecular_biology	PDB_API; FPbase_API	sequence_resources; protein_reference; external_protein_data	single_sequence_artifact	protein_selection; sequence_design	expected_behavior_stated; explicit_requirements; location_supplied	sequence_length; gc_content; sequence_order; external_data_accuracy	sequence_format; size_validation	very hard; mean reward 0.4545; 110 scored trials
43	terminal21	security_testing	web_security; cross_site_scripting	HTML; JavaScript	supplied_program; validation_script	single_output_file	adversarial_security_reasoning	expected_behavior_stated; explicit_requirements; location_supplied	automatic_execution; interaction_free	test_script	hard; mean reward 0.7727; 110 scored trials
44	terminal21	interface_implementation	terminal_emulation; shell_automation	Python; Bash	existing_interface	single_module_implementation	interactive_process_control; terminal_input_handling	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	startup_environment; system_install	importability	hard; mean reward 0.7545; 110 scored trials
45	terminal21	software_build	computational_topology; scientific_computing	Python; Cython; NumPy	existing_package; documentation_snippet; existing_tests	package_build_and_compatibility	compatibility_debugging	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	version_compatibility; test_exclusions	integration_test; test_suite	medium; mean reward 0.8818; 110 scored trials
46	terminal21	document_processing; data_extraction	document_processing; financial_data	unknown	supplied_documents	file_organization; report_generation	document_classification; financial_extraction	expected_behavior_stated; explicit_requirements; location_supplied	field_rule; precedence_rule; input_mutation	output_schema; directory_state	medium; mean reward 0.8807; 109 scored trials
47	terminal21	numerical_optimization	probability; information_theory	Python; NumPy; SciPy	supplied_constraints	distribution_generator	numerical_optimization; constraint_solving	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	numerical_tolerance; vocabulary_size; validity	numerical_validation; artifact_format	easy; mean reward 0.9636; 110 scored trials
48	terminal21	batch_scheduling	machine_learning_systems; batch_optimization	unknown	request_data	plan_artifacts	optimization; constraint_satisfaction	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	shape_alignment; shape_limit; performance_threshold; input_integrity	cost_model; performance_threshold; coverage	easy; mean reward 0.9364; 110 scored trials
49	terminal21	data_inference; causal_modeling	bayesian_networks	unknown	supplied_dataset	data_artifact_generation; model_and_dataset_outputs	statistical_structure_learning; causal_inference	expected_behavior_stated; explicit_requirements; location_supplied	graph_structure; edge_count; intervention_value	artifact_generation; statistical_validation	very easy; mean reward 1.0000; 109 scored trials
50	terminal21	service_implementation; automated_deployment	version_control; web_hosting	Git; SSH; Nginx; HTTPS	local_service	multi_branch_service	branch_deployment; service_integration	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	deployment_latency; branch_separation; self_signed_tls	end_to_end_push; http_test	very easy; mean reward 0.9817; 109 scored trials
51	terminal30	security_bug_fix	binary_security; reverse_engineering; image_conversion	Python	stripped_service_binary; black_box_behavior	patcher_and_binary_artifacts	black_box_behavioral_inference; binary_analysis; generalization	expected_behavior_stated; explicit_requirements; location_supplied	original_binary_unchanged; patch_size_limit; preserve_valid_behavior; no_stub_or_wrapper	behavioral_testing; artifact_verification	very hard; mean reward 0.0000; 60 scored trials
52	terminal30	operational_compliance_workflow	trade_compliance; customs_reporting; data_reconciliation	unknown	SOP_and_reference_filings; portal_error_catalog; service_credentials	multi_system_workflow	data_reconciliation; policy_interpretation; approval_workflow	expected_behavior_stated; explicit_requirements; location_supplied	four_eyes_approval; period_boundary_rules; archive_metadata_requirements	service_setup; submission_confirmation; artifact_verification	very hard; mean reward 0.0000; 60 scored trials
53	terminal30	scientific_calculation	radiological_metrology; food_safety; analytical_chemistry	unknown	measurement_spreadsheet; sample_data; calibration_reference_pdf	calculation_artifact	scientific_reasoning; measurement_uncertainty; numerical_calculation	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	well_known_blank_assumption; significant_figures; exact_report_format	artifact_generation; calculation_validation	hard; mean reward 0.0500; 60 scored trials
54	terminal30	parametric_modeling	computer_aided_design; mechanical_engineering; turbomachinery	FreeCAD; Python	explicit_geometry_specification	script_and_model_artifacts	geometric_reasoning; parametric_design	expected_behavior_stated; explicit_requirements; location_supplied	single_solid; feature_tree_required; parameter_edit_invariant	artifact_generation; artifact_verification	hard; mean reward 0.0333; 60 scored trials
55	terminal30	parameter_optimization	advertising_technology; causal_measurement; online_control	unknown	running_service; OpenAPI_specification; streamed_event_data	operational_configuration	statistical_reasoning; optimization; noise_robustness	expected_behavior_stated; explicit_requirements; location_supplied	minimum_airtime; change_freeze_window; target_metric	service_setup; metric_evaluation	medium; mean reward 0.1500; 60 scored trials
56	terminal30	financial_calculation	counterparty_credit_risk; regulatory_capital; finance	unknown	trade_and_csa_data; regulatory_factor_tables; holiday_calendars	analysis_and_workbook_artifacts	financial_reasoning; regulatory_calculation; formula_auditability	expected_behavior_stated; explicit_requirements; location_supplied	CRR3_rules; fixed_as_of_date; required_rounding; formula_preservation	artifact_generation; artifact_verification; calculation_reconciliation	medium; mean reward 0.2000; 60 scored trials
57	terminal30	scientific_data_analysis	crystallography; hydrogen_bond_networks; topological_graph_analysis	unknown	CIF_structure_files; explicit_geometric_definitions	analysis_artifact	geometric_reasoning; graph_theory; scientific_classification	expected_behavior_stated; explicit_requirements; location_supplied	distance_and_angle_cutoffs; include_both_node_types; exact_output_schema	artifact_generation; numerical_validation	easy; mean reward 0.3167; 60 scored trials
58	terminal30	database_migration	database_systems; web_api; distributed_systems	MySQL; PostgreSQL; Python	running_API_server; baseline_behavior	multi_service_migration	architectural_tradeoffs; concurrency_reasoning; consistency_reasoning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	zero_downtime; no_stale_reads; behavioral_compatibility; latency_budget; future_deploy_state	integration_tests; latency_measurement; consistency_validation	easy; mean reward 0.2833; 60 scored trials
59	terminal30	operational_planning	enterprise_resource_planning; procurement; manufacturing	Odoo	ERP_instance; operating_records; supplier_and_capacity_rules	multi_system_workflow	constraint_satisfaction; procurement_optimization; schedule_planning	expected_behavior_stated; explicit_requirements; location_supplied; environment_supplied	customer_budgets; deadlines; capacity_limits; supplier_limits; traceable_lineage	service_setup; record_validation; artifact_verification	very easy; mean reward 0.6037; 60 scored trials
60	terminal30	artifact_generation	graphic_design; computer_vision; layout_reconstruction	Python	target_layout_image; component_asset_library; renderer	configuration_artifact	visual_inference; geometric_reasoning; layer_composition	expected_behavior_stated; explicit_requirements; location_supplied	pixel_match_threshold; provided_assets_only; editable_text_requirement	rendered_artifact_comparison; pixel_accuracy_measurement	very easy; mean reward 0.6500; 60 scored trials
61	terminal40	parametric_cad_modeling	mechanical_design	FreeCAD_Python	explicit_geometry_parameters	artifact_generation	parametric_modeling	explicit_requirements	structural_artifact_requirements	artifact_verification	very hard; mean reward 0.0074; 135 scored trials
62	terminal40	workflow_cli_implementation	logistics_scheduling	CLI	event_feed_and_operational_packet	whole_program_implementation	stateful_planning	explicit_requirements; location_supplied	state_consistency	evaluation_harness	very hard; mean reward 0.0000; 135 scored trials
63	terminal40	statistical_data_analysis	proteomics	Broad_GSEA_CLI	local_expression_dataset	analysis_and_artifact_generation	statistical_reasoning	explicit_requirements; location_supplied	reproducibility	artifact_generation	hard; mean reward 0.2741; 135 scored trials
64	terminal40	visual_layout_reconstruction	graphic_design_reconstruction	CSS; Python	local_image_asset_library	structured_artifact_generation	visual_reconstruction	explicit_requirements; location_supplied	artifact_constraints	pixel_match_evaluation	hard; mean reward 0.1333; 135 scored trials
65	terminal40	case_decision_workflow	warranty_claims	portal_API	local_policy_and_claim_data	external_workflow	policy_application	explicit_requirements; location_supplied	authoritative_local_policy	external_workflow	medium; mean reward 0.4370; 135 scored trials
66	terminal40	regulatory_filing_workflow	trade_statistics_compliance	API	local_sop_and_reference_records	external_workflow	reconciliation_and_audit	explicit_requirements; location_supplied	filing_completion	external_workflow	medium; mean reward 0.4074; 135 scored trials
67	terminal40	algorithm_implementation	cryptography	SageMath_Python	local_cipher_resources	whole_program_implementation	cryptanalysis	explicit_requirements; location_supplied	performance	artifact_generation	easy; mean reward 0.5704; 135 scored trials
68	terminal40	cryptanalysis_tool	classical_cryptography	Python	ciphertext_input	whole_program_implementation	cryptanalysis	explicit_requirements; location_supplied	performance	behavioral_correctness	easy; mean reward 0.5926; 135 scored trials
69	terminal40	formal_proof	formal_verification	Coq	formal_project_context	local_code_change	formal_proof_reasoning	explicit_requirements; location_supplied	formal_soundness	build_checking	very easy; mean reward 0.8370; 135 scored trials
70	terminal40	model_checkpoint_conversion	machine_learning_checkpoints	Safetensors_HuggingFace	parallel_shards_and_reference_outputs	artifact_generation	tensor_parallel_reconstruction	explicit_requirements; location_supplied	compatibility	artifact_verification	very easy; mean reward 0.8444; 135 scored trials
71	toolathlon	analysis	data_analysis_and_statistics	Excel	workspace_files; email_context; existing_external_records; available_tool_excel; available_tool_emails	workspace_artifact_edit; external_communication	numerical_analysis; information_extraction	explicit_requirements; output_location_supplied; environment_or_service_supplied	threshold_or_filter	artifact_generation; external_communication	very hard; mean reward 0.0891; 101 scored trials
72	toolathlon	information_retrieval; artifact_or_record_creation; planning	education_administration	Excel	workspace_files; available_tool_pdf_tools; available_tool_excel	workspace_artifact_edit	information_extraction; planning_or_optimization	output_format_specified; explicit_requirements	scope_restriction; date_or_time_window	artifact_generation	very hard; mean reward 0.0000; 101 scored trials
73	toolathlon	artifact_or_record_creation	academic_research	GitHub; Git	email_context; available_tool_emails; available_tool_github	external_record_or_resource_change	information_extraction; source_research	environment_or_service_supplied	unknown	external_record_or_resource_change	hard; mean reward 0.1961; 102 scored trials
74	toolathlon	analysis; communication; planning	business_operations	WooCommerce	email_context; available_tool_woocommerce; available_tool_emails	external_communication	planning_or_optimization	environment_or_service_supplied	threshold_or_filter; date_or_time_window	external_communication	hard; mean reward 0.1700; 100 scored trials
75	toolathlon	analysis; artifact_or_record_creation	financial_analysis	Excel	workspace_files; available_tool_excel	workspace_artifact_edit	numerical_analysis	output_location_supplied	source_restriction	artifact_generation	medium; mean reward 0.4118; 102 scored trials
76	toolathlon	information_retrieval	business_operations	Google_Sheets; Kubernetes	workspace_files; available_tool_k8s; available_tool_google_sheet	workspace_artifact_edit	information_extraction	explicit_requirements; environment_or_service_supplied	unknown	unknown	medium; mean reward 0.3861; 101 scored trials
77	toolathlon	information_retrieval; artifact_or_record_creation	media_analysis	Notion; YouTube	existing_external_records; external_research; available_tool_playwright_with_chunk; available_tool_notion; available_tool_fetch; available_tool_web_search	external_record_or_resource_change	information_extraction; source_research	environment_or_service_supplied	unknown	external_record_or_resource_change; requested_completion_check	easy; mean reward 0.5545; 101 scored trials
78	toolathlon	analysis	personal_productivity	unknown	workspace_files; email_context; existing_external_records; available_tool_emails; available_tool_snowflake; available_tool_pdf_tools	external_communication	information_extraction	output_format_specified	source_restriction; threshold_or_filter	external_communication	easy; mean reward 0.4804; 102 scored trials
79	toolathlon	information_retrieval	academic_research	arXiv	workspace_files; external_research; available_tool_arxiv_local; available_tool_scholarly	unknown	information_extraction; source_research	explicit_requirements	scope_restriction; date_or_time_window	requested_completion_check	very easy; mean reward 0.7941; 102 scored trials
80	toolathlon	analysis; artifact_or_record_creation	software_and_data_tools	GitHub; Git	workspace_files; available_tool_fetch; available_tool_web_search	workspace_artifact_edit	unknown	output_format_specified; output_location_supplied; environment_or_service_supplied	unknown	artifact_generation	very easy; mean reward 0.9314; 102 scored trials
81	verified	bug_report_investigation	data_visualization	Matplotlib; Python; LaTeX; pandas	datetime_axis_reproduction	local_edit	rendered_spacing_comparison	expected_output_comparison	version_comparison	rendering_reproduction	very hard; mean reward 0.0000; 15 scored trials
82	verified	API_consistency_fix	scientific_data_processing	xarray; Python	DataArray_and_Dataset_API_comparison	public_API_parameter_change	coordinate_vs_dimension_semantics	expected_behavior_stated	preserve_coordinate_semantics	unknown	very hard; mean reward 0.0000; 15 scored trials
83	verified	feature_implementation	database_storage	Django; Python	supplied_reproduction; existing_behavior_description	local_edit	data_model_reasoning	expected_behavior_stated; explicit_requirements	unknown	reproduction_check; expected_result_check	hard; mean reward 0.0667; 15 scored trials
84	verified	bug_fix	symbolic_mathematics; code_generation	SymPy; Python; Unicode	supplied_snippet	local_edit	unicode_rendering_reasoning	expected_behavior_stated	subscript_formatting	unknown	hard; mean reward 0.0667; 15 scored trials
85	verified	feature_implementation	web_application_frameworks	Django; PostgreSQL	database_configuration_context	public_configuration_change	configuration_propagation	explicit_requirements; expected_behavior_stated	mutual_tls_context	unknown	medium; mean reward 0.2000; 15 scored trials
86	verified	bug_fix	networking; date_time_processing	Django	standard_requirement_provided; current_behavior_described	component_change	date_calculation_reasoning	expected_behavior_stated	RFC_7231_compatibility	expected_result_check	medium; mean reward 0.1333; 15 scored trials
87	verified	bug_fix	data_format_registry	Python; Astropy; HENDRICS	error_trace; reproduction_steps	library_source_change	root_cause_analysis; regression_compatibility_analysis	failing_behavior_stated; reproduction_provided	unknown	regression_testing	easy; mean reward 0.4000; 15 scored trials
88	verified	feature_request	scientific_data_processing	xarray; Python	multi_dimension_reduction_example	local_edit	valid_element_count_across_dimensions	explicit_requirements; expected_behavior_stated	min_count_multi_dimension_case	unknown	easy; mean reward 0.3333; 15 scored trials
89	verified	behavior_change	database_storage	Django; Python	code_example_supplied; version_behavior_described	component_change	compatibility_reasoning; database_state_reasoning	expected_behavior_stated; example_provided	backward_compatibility	reproduction_example; expected_result_check	very easy; mean reward 0.5333; 15 scored trials
90	verified	bug_fix	command_line_utilities; documentation_tooling	Sphinx; Python	existing_codebase	local_edit	interactive_cli_reasoning	expected_behavior_stated	exit_behavior	behavioral_testing	very easy; mean reward 0.5333; 15 scored trials

90 difficulty-balanced tasks: ten axes