Hugging Face's logo

Datasets:
achinta3
/
cybersec-jsonschemabench

Tasks:
Text Generation
Modalities:
Tabular
Text
Formats:
json
Size:
< 1K
Tags:
cybersecurity
json
structured-outputs
long-context
reasoning
benchmarking
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
unique_id
stringlengths
56
84
subset
stringclasses
2 values
split
stringclasses
1 value
task_type
stringclasses
9 values
prompt
stringclasses
9 values
json_schema
stringclasses
1 value
input_jsonl
stringlengths
820k
873k
ground_truth_json
stringlengths
839
9.77k
answer_schema_name
stringclasses
1 value
input_record_count
int64
500
500
record_start
int64
0
5.8k
record_end
int64
500
6.3k
slice_jsonl_path
stringlengths
160
216
score_fields
listlengths
1
1
difficulty_tags
listlengths
6
6
reasoning_family
stringclasses
9 values
candidate_count
int64
1
11
distractor_count
int64
0
347
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.000
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
4,650
5,150
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
4
184
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.001
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
3,300
3,800
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
4
121
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.002
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
5,500
6,000
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
3
150
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.003
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
3,250
3,750
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
3
62
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.004
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
5,000
5,500
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
2
158
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.005
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
4,950
5,450
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
2
145
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.006
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
5,700
6,200
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
2
103
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.007
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
4,850
5,350
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
2
95
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.008
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
4,750
5,250
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
2
95
flaws_cloudtrail.hard.cloudtrail_request_parameter_event_principal_symdiff_rank1.009
flaws_cloudtrail
test
hard_cloudtrail_request_parameter_event_principal_symdiff_rank1
"Use only the 500-record CloudTrail JSONL slice. Do not start from a named anchor. Across these nest(...TRUNCATED)
"{\n \"$schema\": \"https://json-schema.org/draft/2020-12/schema\",\n \"$id\": \"https://jsonsecur(...TRUNCATED)
"{\"benchmark_source\": \"flaws_cloudtrail\", \"source_file\": \"flaws_cloudtrail_logs/flaws_cloudtr(...TRUNCATED)
"{\"task_type\":\"multi_hop_record_answer\",\"dataset\":\"flaws_cloudtrail\",\"reasoning_type\":\"ra(...TRUNCATED)
multi_hop_record_answer
500
4,900
5,400
"artifacts/cybersec_jsonschemabench_hard_v1/slices/flaws_cloudtrail/cloudtrail_request_parameter_eve(...TRUNCATED)
[ "answer_record_ids" ]
["unanchored_selection","ranked_selection","nested_request_parameters","dual_projection","symmetric_(...TRUNCATED)
cloudtrail_request_parameter_event_principal_symdiff_rank1
2
89
End of preview. Expand in Data Studio

CybersecJSONSchemaBench Hard

This hard split is a JSONSchemaBench-style cybersecurity benchmark built from normalized CloudTrail and Suricata EVE records. It replaces anchored lookup questions with unanchored, deterministic multi-hop reasoning programs over large nested JSONL slices.

Each row includes:

  • unique_id
  • json_schema
  • prompt
  • input_jsonl
  • ground_truth_json
  • reasoning_family
  • candidate_count
  • distractor_count

Current Version

  • benchmark version: 1.0.0-hard
  • total rows: 100
  • window size: 500 JSONL records
  • split: test
  • sources: flaws_cloudtrail, suricata_eve

Evaluation

The local scorer validates model output against multi_hop_record_answer and scores exact ordered answer_record_ids. Gold answers are generated by deterministic task programs and then recomputed from each serialized slice.
Downloads last month
61
Number of rows:
100
Total file size:
94.3 MB