Datasets:

achinta3
/

cybersec-jsonschemabench-cloudtrail-objective-natural-hard-v5

instance_id string	benchmark_name string	benchmark_version string	dataset string	task_type string	prompt string	answer_schema dict	input_jsonl string	input_artifact dict	evaluation dict	reasoning_family string	difficulty_tags list	candidate_count int64	distractor_count int64	primary_margin_over_runner_up int64	evidence_dispersion_records int64	input_jsonl_bytes int64
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.000	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Identify the identity with the largest discovery-before-change progression in the CloudTrail data, (...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	3	1,444	31	823	3,058,801
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.001	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Which identity shows the largest progression from reconnaissance to change? Compare identities by d(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	3	1,309	4	687	3,216,304
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.002	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Find the identity whose activity most extensively moves from discovery into state-changing behavior(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	3	1,038	15	1,209	3,201,340
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.003	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Determine the identity with the largest discovery phase preceding its first state-changing action. (...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	3	881	6	832	3,500,766
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.004	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"From the CloudTrail activity, identify the identity with the largest discovery-before-change progre(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	2	1,558	9	1,503	2,997,979
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.005	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Which identity had the broadest discovery activity before making a state-changing API call? Apply c(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	2	1,520	8	1,166	3,608,208
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.006	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Identify the identity that progresses furthest from discovery APIs into mutation APIs, comparing id(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	2	1,502	30	806	3,057,190
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.007	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Find the identity with the largest reconnaissance-to-change sequence and cite the earliest supporti(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	2	1,479	21	1,208	3,153,126
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.008	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Which identity has the largest number of distinct discovery API actions before its first state-chan(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	2	1,471	146	1,362	3,051,255
flaws_cloudtrail.natural_hard_v5.actor_recon_to_change.009	CybersecJSONSchemaBench	5.0.0-cloudtrail-objective-natural-hard	flaws_cloudtrail	multi_hop_record_answer	"Find the activity trail where one identity performs the widest discovery phase before changing clou(...TRUNCATED)	{"$schema":"https://json-schema.org/draft/2020-12/schema","$id":"https://jsonsecuritybench.local/sch(...TRUNCATED)	"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)	{"path":"artifacts/cybersec_jsonschemabench_demo/raw/flaws_cloudtrail_logs.tar","record_count":1600,(...TRUNCATED)	{"primary_metric":"schema_validity_gated_ordered_id_exact","score_fields":["answer_record_ids"],"fie(...TRUNCATED)	actor_recon_to_change	["temporal_ordering","read_then_write","cross_service","objective_natural_prompt","long_context_1600(...TRUNCATED)	2	1,467	294	977	3,094,169

End of preview. Expand in Data Studio

CybersecJSONSchemaBench CloudTrail Objective Natural Hard v5

This is a 100-problem natural-prompt long-context cybersecurity reasoning subset built from the full flAWS CloudTrail corpus.

Each row contains a natural analyst-style question, a large CloudTrail JSONL context, and the JSON schema the answer must match. Gold answers are deterministic hidden-oracle results over the serialized slice and are not included in this public export.

Families

actor_recon_to_change: 22
apigateway_mapping_chain: 8
credential_consequence_chain: 16
cross_service_value_bridge: 4
ec2_unsuccessful_expansion: 30
s3_asset_pressure: 4
shared_client_divergence: 16

Columns

instance_id: Stable problem identifier.
prompt: Natural benchmark-style analyst question.
answer_schema: JSON schema for the expected answer.
input_jsonl: Serialized CloudTrail records for the problem context.
evaluation: Scoring metadata.
reasoning_family: Hidden-oracle family name.
difficulty_tags: Intended reasoning requirements.
candidate_count: Number of competing candidates under the hidden oracle.
distractor_count: Candidate-related records not in the gold answer.
primary_margin_over_runner_up: Dominance margin used to control ambiguity.
input_jsonl_bytes: Serialized context size.

Downloads last month: 11