solsticestudioai/nemesis-cyber-pack

Nemesis Cyber Threat Simulation Pack (Sample)

A synthetic adversarial-agent cyber operations dataset for detection-model training, SOC analyst triage research, and blue-team evaluation. Each row captures a complete simulated attack episode: triggering anomaly, environment context, adversarial planner reasoning, correlated telemetry trace, execution summary, and final decision outcome (detected / blocked / impact achieved / stealth maintained / exfiltration complete).

Built by SolsticeAI as a free sample of a larger commercial pack. 100% synthetic. No real incident, victim, or exploit data — and no working offensive code. TTP labels align with MITRE ATT&CK vocabulary so this sample can be used to train and benchmark defenders.

What is included

File	Rows	Format	Purpose
nemesis_cyber_sample.parquet	10,000	Parquet	Columnar, typed, best for analytics
nemesis_cyber_sample.jsonl	10,000	JSON Lines	Streaming / LLM training friendly

Source pack: 2.5M-episode corpus
This sample: 10,000 episodes, stratified 2,000 per outcome class
Outcome classes: detected_by_soc, blocked_by_edr, stealth_maintained, exfiltration_complete, impact_achieved
Environments covered: AWS-Cloud, Active-Directory, Kubernetes, Web-App-Gateway

Record structure

Each record is one simulated attack episode with 8 top-level fields:

Field	Type	Contents
schema_version	string	Pack schema version (1.0.0-nemesis-cyber-sample)
event	struct	id, timestamp, trace_id, weighted_score, decision_outcome
risk_context	struct	trigger, protocol, chain, impacted_asset, anomaly_signature
agent_reasoning	struct	engine, winning_strategy, confidence_score, mcts_branches
correlated_telemetry	list	Ordered action chain with per-step telemetry (latency, noise, evasion score, node provider)
execution_summary	struct	strategy, success_rate, total_execution_ms, noise_penalty
genetic_optimizer_feedback	struct	fitness_score_update, parameter_drift
decision_outcome	string	Final label (duplicated from event.decision_outcome for convenience)

See SCHEMA.md for the full nested field breakdown.

Why this dataset is useful

Most public cybersecurity datasets are either raw packet captures, static CTI feeds, or narrow single-technique labeling sets. This pack is shaped around what detection-engineering and SOC-analytics teams actually need to train modern models:

• Multi-step attack episodes rather than isolated alerts
• Balanced outcome classes across detected, blocked, stealthy, and successful attempts
• Adversarial reasoning trace (strategy + MCTS branch count + confidence) alongside the telemetry
• Per-step evasion and noise signals to train detection models that weigh stealth vs noise trade-offs
• Cross-environment coverage (cloud, identity, container, web)
• Stable schema suitable for dashboard prototyping, triage simulators, and ML pipelines

Typical use cases

• SOC triage and alert-prioritization model training
• Detection engineering rule evaluation against balanced positive and negative cases
• Adversarial-AI research on multi-step planner behavior
• Tabletop and red-vs-blue simulator content
• LLM fine-tuning on incident narratives and defender reasoning
• Benchmarking anomaly-scoring and false-positive reduction pipelines
• Dashboard and BI template development for security analytics

Quick start

import pandas as pd
import pyarrow.parquet as pq

df = pq.read_table("nemesis_cyber_sample.parquet").to_pandas()

# Outcome distribution (stratified balanced)
print(df["decision_outcome"].value_counts())

# Evasion pressure per environment
df["protocol"] = df["risk_context"].apply(lambda r: r.get("protocol"))
df["avg_evasion"] = df["correlated_telemetry"].apply(
    lambda steps: sum(s["telemetry"]["evasion_score"] for s in steps) / max(len(steps), 1)
)
print(df.groupby("protocol")["avg_evasion"].mean().round(3))

# Detection-rate by trigger type
df["trigger"] = df["risk_context"].apply(lambda r: r.get("trigger"))
detection_rate = (df["decision_outcome"].isin(["detected_by_soc", "blocked_by_edr"])
                  .groupby(df["trigger"]).mean().round(3))
print(detection_rate)

Streaming form:

import json

with open("nemesis_cyber_sample.jsonl") as f:
    for line in f:
        episode = json.loads(line)
        # one episode per line

Responsible use

This dataset is intended for defensive research: detection modeling, SOC tooling, and adversarial-agent studies. It contains synthesized attack metadata and MITRE-aligned TTP labels — it does not contain working offensive payloads, exploit code, shellcode, malware samples, credentials, private vulnerability details, or any real-world victim data. Please use it to improve defenses.

License

Released under CC BY 4.0. Use freely for research, detection-engineering, education, and commercial prototyping with attribution.

Get the full pack

This Hugging Face repo is a 10K-episode sample. The production pack scales to 2.5M+ episodes, additional outcome labels, richer per-step telemetry, attacker/defender variant splits, multi-environment campaign chains, parquet + JSONL + SIEM-import formats, and buyer-specific variants.

Self-serve (Stripe checkout):

• Sample Scale tier — $5,000 — ~25K records, one subject, 72-hour delivery.

Full pack + enterprise scope:

• www.solsticestudio.ai/datasets — per-SKU pricing across Starter / Professional / Enterprise tiers, plus commercial licensing, custom generation, and buyer-specific variants.

Procurement catalog:

• SolsticeAI Data Storefront — available via Datarade / Monda.

Citation

@dataset{solstice_nemesis_cyber_pack_2026,
  title        = {Nemesis Cyber Threat Simulation Pack (Sample)},
  author       = {SolsticeAI},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/solsticestudioai/nemesis-cyber-pack}
}