the-mitre-corporation/Wizard_Spider_Cyberattack_Timeline_Reconstruction_Dataset

Wizard Spider Cyberattack Timeline Reconstruction Dataset v1.0

Dataset description

This dataset supports reconstruction of cyberattack activities and their predecessor relationships from Windows event logs and network packet captures. It contains telemetry from one controlled Wizard Spider-inspired emulation, an analyst-reviewed reference comprising 36 aggregate activities and 313 supporting observations, and the EL attack graph used for emulation. Evaluation implementation, or example evaluation results are not included in this release.

The dataset contains the following:

1. Endpoint logs and network captures from a controlled, Wizard Spider-inspired attack emulation. This emulation was inspired by MITRE ATT&CK Evaluations' Wizard Spider Scenario 1 plan. Benign user activity and background traffic are included alongside attack activity.
2. An analyst-reviewed reference timeline of the attack supported by the collected logs.
3. The attack graph containing attack commands, used in the emulation, is also provided; its purpose is only to help understand the reference timeline. Any tool attempting to generate an attack reconstruction is cautioned not to use the attack graph to guide reconstruction, to prevent bias.

The package is a collection of related research artifacts with different schemas. The Hugging Face viewer is disabled to preserve this organization; see Hugging Face's viewer configuration documentation.

Publisher and dataset authors

Dataset publisher: The MITRE Corporation, USA.

Dataset authors, in submission order:

1. Suresh K. Damodaran
2. Rebecca Steinberg
3. Caleb Scopetski
4. Victor H. Paiz
5. Jared J. Digles
6. Robert Mentkowski
7. Hayden Mcdonald
8. Amey Joshi
9. Alondra Granados

Supported tasks and languages

Intended research tasks include:

• Attack timeline reconstruction from endpoint and network logs by third party tools.
• Evaluating attack reconstructions against the reference timeline provided.

Repository contents

README.md
LICENSE
attack_graph/
  ws3.el.json
dataset/
  Wizard-Spider-12-18.evtx
  Wizard_Spider-12-18.pcapng
reference/
  reference_events.json
  reference_observations.json

Artifact	Contents
Endpoint capture	Windows event-log capture used as Sysmon evidence in the case study.
Network capture	Packet capture used to derive Zeek network records.
Aggregated Reference	Analyst-reviewed logical attack events, collected evidence values, severity labels, and predecessor relationships. Referred to as "reference timeline" in this document.
Observation reference	Flattened observation records associated with each aggregate ID in the reference timeline.
EL attack graph	Executable emulation specification with graphs, nodes, edges, watchpoints, effects, and embedded commands/scripts.

Data structure and fields

Units of data

An observation records evidence derived from telemetry. An aggregate reference event represents a logical attack activity supported by one or more observations. Several observations can support the same reference event.

Aggregate reference events

reference/reference_events.json is a JSON object with an events array containing 36 records.

Field	JSON type	Meaning
id	integer	Logical reference-event identifier.
ts	array of strings	Collected observation timestamps in UTC.
raw_ids	array of strings	Source telemetry identifiers collected for verification.
hosts	array of strings	Hostnames and/or IP addresses associated with the activity.
destinations	array of strings	Destination entities collected from the evidence.
datasets	array of strings	Evidence-source labels such as sysmon, zeek.http, or zeek.conn.
summary	array of strings	Collected descriptions of the aggregate activity.
severity	string	Analyst-assigned severity: Low, High, or Critical in this reference.
correlations	array of objects	Predecessor aggregate IDs and relationship rationales, using id and pivots.
command	string	Attack command or analytical description of the step.

The arrays retain aggregate evidence values and must not be assumed to be parallel arrays that can be zipped into raw records. Use the observation reference for the packaged row representation.

Reference observations

reference/reference_observations.json contains source, grouping, and an events array of 313 records. Its grouping value is events_split_to_one_timestamp_raw_id_pair; source records an upstream filename, not another included file.

Field	JSON type	Meaning
id	integer	Observation identifier within this reference file.
aggregate_id	integer	Logical reference-event membership.
ts	string	UTC timestamp; precision varies and may exceed microseconds.
raw_id	string	Source telemetry record identifier.
host, destination, dataset	string	Scalar entity and source values for the observation.
summary	string or array of strings	Description(s) of the supporting observation; may be more specific than the aggregate activity.
severity	string	Reference severity label.
aggregate_correlations	array of objects	Predecessor aggregate-event relationships, using aggregate_id and pivots.
command	string	Command or activity description.

Observation records use singular raw_id, host, destination, and dataset fields; the aggregate reference retains plural field names. Observations are unique by (aggregate_id, dataset, ts, raw_id). Telemetry identifiers can recur across evidence sources and aggregate events. A summary array contains multiple descriptions of the same observation.

For example, the first packaged observation is:

{
  "id": 1,
  "aggregate_id": 1,
  "ts": "2024-12-18T21:10:40.040241Z",
  "raw_id": "C9al3f12QZLVFPLf77",
  "host": "192.168.1.7",
  "destination": "192.168.0.4",
  "dataset": "zeek.http",
  "summary": "Dorothy maintained WinHTTP-based C2 polling and output traffic with the attacker server using /, /getTask, and /output during the Judy/adb payload phase.",
  "severity": "Low",
  "aggregate_correlations": [],
  "command": "Dorothy WinHTTP C2 beaconing and task/output exchange"
}

Dataset statistics

Component	Verified size or count
Raw EVTX file	634,458,112 bytes; 139,869 event records
Raw PCAPNG file	255,727,780 bytes; 277,390 packets
Aggregate reference events	36
Flattened reference observations	313
Unique direct reference causal edges	39
Reference transitive reachable pairs, excluding self-relations	204
EL attack graph	82 nodes; 95 edges

Raw packet counts are not Zeek log counts. Processed Zeek logs are not separately packaged here; the file sizes and record counts above describe the included raw captures and reference artifacts.

Reference severity	Aggregate events	Flattened observations
Low	6	130
High	20	130
Critical	10	53
Total	36	313

The flattened reference source distribution is zeek.http 188, sysmon 100, zeek.conn 6, zeek.ssh 7, zeek.smb_files 5, zeek.files 4, zeek.smb_mapping 2, and zeek.kerberos 1. Reference severities describe the curated attack activities; the raw captures do not provide exhaustive severity labels for every benign and malicious record.

The collected reference timestamps range from 2024-12-18T21:10:40.040241Z to 2024-12-18T21:16:21.366065Z. The packet capture spans 2024-12-18T21:10:32.811734Z to 2024-12-18T21:16:29.280536Z.

Dataset creation

Source and collection

The raw logs were collected from an adversary emulation in a cyber range. The EL based emulation approach is described in Suresh K. Damodaran and Paul D. Rowe's paper Automated Repeatable Adversary Threat Emulation with Effects Language (EL). That work explains executable attack graphs, and Wizard Spider plan translation.

Synthetic normal users generated background activity before, during, and after the emulation across accounts and victim hosts using the GHOSTS NPC Framework. Endpoint events and network traffic, including DNS and HTTP activity, were captured. Network packets were processed with Zeek for analysis and indexed alongside Sysmon records in a containerized SIEM with a two-node Elasticsearch cluster, Filebeat, and Kibana.

Reference construction and annotation

Reference construction used ten independent GPT-5.5 investigation sessions with no shared chat context. The attack graph, embedded commands, investigative pivots, and endpoint/network telemetry guided the investigations. Reference construction was guided by the known attack plan in the provided EL graph.

Observations were aggregated around an attack command or step on a particular asset. GPT-5.5 assisted in aggregation and summary generation. An analyst reviewed every reference event and causal relationship before approving the reference for scoring. Severity labels were assigned to aggregate events. Some asynchronous activities, such as beaconing, may not be connected to other events.

Intended uses, biases, and limitations

The dataset is intended for defensive research, method development, forensic education, threat hunting education, and transparent evaluation of structured attack reconstructions.

Privacy, sensitive content, and ethical considerations

The raw logs were collected from an adversary emulation in a cyber range. Background simulated user activity and network payloads are also present.

The artifacts retain laboratory created synthetic hostnames, account names, private IP addresses, file paths, commands, URLs, and telemetry identifiers. Credential-bearing strings intentionally occur in the executable attack graph, reference commands/summaries, and supporting observations.

The EL graph is an executable attack-emulation artifact with offensive commands/scripts. We do not recommend third party tools using the attack-emulation artifacts to guide the generation of the attack reconstruction.

Licensing

This dataset, including the EVTX and PCAPNG captures, reference annotations, and attack graph, is released under the Apache License, Version 2.0. See LICENSE for the full terms and the Apache Software Foundation's license text.

The upstream Wizard Spider plan/library also identifies Apache License 2.0 terms; see its license file. Applicable upstream attribution and notices remain relevant to reused material.

Access and reproducibility

Download the repository files as a snapshot. The Hugging Face snapshot-download API supports dataset repositories and revision pinning. No unified load_dataset() builder is provided for this mixed binary/JSON artifact package.

Run the following from the downloaded repository root to inspect the references using Python's standard library:

import json
from pathlib import Path

root = Path(".")
reference_events = json.loads(
    (root / "reference/reference_events.json").read_text(encoding="utf-8")
)["events"]
reference_observations = json.loads(
    (root / "reference/reference_observations.json").read_text(encoding="utf-8")
)["events"]

print("Reference events:", len(reference_events))
print("Reference observations:", len(reference_observations))

The EVTX file requires an event-log parser; the PCAPNG file requires a packet-analysis tool. Recreating the SIEM evidence also requires Zeek processing and the original indexing/normalization rules. These tools and their exact versions are not bundled.

Pin the repository revision and record the input-file hashes for any downstream evaluation.

Citation

Please cite the dataset as below. The published EL automation paper is cited separately below.

@dataset{damodaran2026wizardspider,
  title = {Wizard Spider Cyberattack Timeline Reconstruction Dataset},
  author = {Damodaran, Suresh K. and Steinberg, Rebecca and
            Scopetski, Caleb and Paiz, Victor H. and Digles, Jared J. and
            Mentkowski, Robert and Mcdonald, Hayden and Joshi, Amey and
            Granados, Alondra},
  year = {2026},
  publisher = {The MITRE Corporation},
  url = {https://huggingface.co/datasets/sdamodaran-mitre/Wizard_Spider_Cyberattack_Timeline_Reconstruction_Dataset}
}

@article{damodaran2026effects,
  title = {Automated Repeatable Adversary Threat Emulation with Effects Language ({EL})},
  author = {Damodaran, Suresh K. and Rowe, Paul D.},
  journal = {Digital Threats: Research and Practice},
  year = {2026},
  doi = {10.1145/3816043},
  url = {https://doi.org/10.1145/3816043}
}

Maintenance and versioning

Dataset contacts:

• Suresh K. Damodaran: sdamodaran@mitre.org
• Rebecca Steinberg: rsteinberg@mitre.org

Future changes to evidence, annotations, model provenance, or scoring should be identified by a new repository revision and accompanying release notes.

©2026 The MITRE Corporation. All rights reserved.

Approved for Public Release; Distribution Unlimited. Public Release Case Number 26-2253.