THULab/cyb011_sample

CYB011 — Synthetic AI Evasion Attack Trajectories (Sample) (TsFile)

Apache TsFile version of xpertsystems/cyb011-sample.

Overview

CYB011 — Synthetic AI Evasion Attack Trajectory Dataset (Sample) — This is a free preview of the full CYB011 — Synthetic AI Evasion Attack Trajectory Dataset product. It contains roughly ~4% of the full dataset at identical schema, attacker-tier distribution, and statistical fingerprint, so you can evaluate fit before licensing the full product.

• Tables: attack_trajectories, campaign_events, campaign_summary, network_topology
• Rows: 27,710 across 4 TsFile table(s).
  • attack_trajectories: 14,000 rows — Time is the source timestep simulation step counter, stored as milliseconds (1 step = 1 ms).
  • campaign_events: 13,310 rows — Time is the source timestep simulation step counter, stored as milliseconds (1 step = 1 ms).
  • campaign_summary: 200 rows — Time is synthetic row index over the stable source order (0 … N-1); the source table has no per-row clock.
  • network_topology: 200 rows — Time is synthetic row index over the stable source order (0 … N-1); the source table has no per-row clock.

Schema (TsFile structure)

attack_trajectories.tsfile

• Time (INT64, milliseconds) — the source timestep simulation step counter, stored as milliseconds (1 step = 1 ms).
• TAG: campaign_id — device dimension(s); query one device with WHERE campaign_id=....
• FIELD (11): attacker_id (string), attack_phase (string), perturbation_magnitude (double), feature_delta_l2_norm (double), feature_delta_linf_norm (double), detector_confidence_score (double), detection_outcome (string), query_count_cumulative (int64), evasion_budget_consumed (double), target_segment_id (string), attacker_capability_tier (string)

campaign_events.tsfile

• Time (INT64, milliseconds) — the source timestep simulation step counter, stored as milliseconds (1 step = 1 ms).
• TAG: campaign_id — device dimension(s); query one device with WHERE campaign_id=....
• FIELD (5): attacker_id (string), event_type (string), target_segment_id (string), detector_id (string), partner_attacker_id (string)

campaign_summary.tsfile

• Time (INT64, milliseconds) — synthetic row index over the stable source order (0 … N-1); the source table has no per-row clock.
• TAG: campaign_id — device dimension(s); query one device with WHERE campaign_id=....
• FIELD (24): attacker_id (string), attacker_capability_tier (string), target_environment_type (string), campaign_type (string), total_queries_issued (int64), evasion_attempts_total (int64), evasion_successes (int64), evasion_success_rate (double), detection_rate (double), perturbation_budget_consumed (double), avg_perturbation_magnitude (double), max_perturbation_linf (double), campaign_duration_steps (int64), lateral_pivot_count (int64), detector_blind_spots_mapped (int64), concept_drift_events (int64), retrain_triggers_caused (int64), campaign_success_flag (int64), attribution_risk_score (double), stealth_score (double), attack_efficiency_score (double), model_extraction_queries (int64), transfer_attack_count (int64), coordinated_attack_flag (int64)

network_topology.tsfile

• Time (INT64, milliseconds) — synthetic row index over the stable source order (0 … N-1); the source table has no per-row clock.
• TAG: segment_id — device dimension(s); query one device with WHERE segment_id=....
• FIELD (11): detector_id (string), segment_type (string), trust_level (double), detection_coverage (double), defender_architecture (string), feature_space_dim (int64), alert_threshold (double), retraining_cadence_days (int64), ensemble_size (int64), detection_strength (double), adversarial_robustness (double)

Conversion notes

• Each CSV of the sample is converted into its own TsFile table (4 tables), so no table is merged with another and no column is dropped.
• Event/telemetry tables use the source simulation clock as Time (INT64 milliseconds): timestep step counters and millisecond offsets are stored as-is, minute-based columns are multiplied by 60,000, and ISO/UTC strings are parsed to epoch milliseconds.
• Static inventory/topology/summary tables have no per-row clock; their Time is a synthetic row index (0..N-1) over the stable source order.
• Low-cardinality dimensions of each table (campaign/incident/asset/session ids, phases, labels) are declared as TAG columns; query one simulation with e.g. WHERE campaign_id='...'.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("attack_trajectories.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: xpertsystems/cyb011-sample
• Source card title / publisher: CYB011 — Synthetic AI Evasion Attack Trajectories (Sample)
• License: cc-by-nc-4.0 (as declared by the original dataset).