Datasets:
Request access to BEACON
Complete this form and agree to the terms to request access.
BEACON contains behavioural biometric data from 28 consenting participants. The screen recordings are not redacted: they can show player names, including the names of other players in a match, and other personal information on screen. Access is for non-commercial research only.
Log in or Sign Up to review the conditions and access this dataset content.
BEACON: Behavioral Engine for Authentication and Continuous Monitoring
An anonymized multimodal behavioral biometric dataset collected during VALORANT gameplay, designed for continuous user authentication research.
28 participants · 78 sessions · 426 GB · release 2.1.1
Participants are represented by opaque IDs (P001–P028) and sessions by sequential
IDs (S001, S002, ...). Raw participant names, consent files, hostnames, MAC
addresses, and direct device IDs are not included in this release.
Access
Access requires an approved request. The screen recordings are not redacted, and the packet captures can reveal network activity, so the dataset is shared under the terms in the request form: non-commercial research only, no attempt to identify participants or other players, and no redistribution of the screen recordings.
Modalities
| Modality | Files | Description |
|---|---|---|
mouse_*.csv |
79 | Mouse telemetry — position, speed, clicks, scroll; Unix epoch timestamps |
keyboard_*.csv |
79 | Keyboard HID dynamics — key identity, press/release times, duration; Unix epoch timestamps. Text typed outside play (log-in forms, browsers, chat) is masked as <letter>, <digit>, or <symbol> |
captured_packets_*.pcap |
79 | The game's own network traffic: UDP to and from the VALORANT game servers (ports 7000–8000 and 8180–8181). IP/MAC addresses are pseudonymized; DNS, TLS/QUIC, other applications' traffic, and local-network discovery traffic are removed. The game payload is encrypted and was not decrypted. |
screen_record_*.mp4 |
73 | Screen recordings (H.264, no audio). Not redacted: they can show player names and other personal information on screen. |
hardware_info_*.json |
78 | Scrubbed hardware metadata (hostname, MAC, IP, serial removed) |
Files with a _part02 suffix belong to the same session as the file without it. In P002/S001
and P009/S002, the screen recording is split into two files. P006/S001 holds two recordings
(19 March and 14 April 2025): its mouse, keyboard, and PCAP files have a _part02 file for the
second recording, which is why there are 79 of each for 78 sessions. Both recordings'
elapsed-time clocks start at 0, so process the two parts separately. Seven sessions have no
screen recording.
File Layout
participant_P001/
session_S001/
mouse_P001_S001.csv
keyboard_P001_S001.csv
captured_packets_P001_S001.pcap
screen_record_P001_S001.mp4
hardware_info_P001_S001.json
Metadata
| File | Description |
|---|---|
metadata/public_participant_session_manifest.csv |
Participant and session ID index |
metadata/public_release_manifest.csv |
Per-file modality, status, size, and SHA-256 |
metadata/sha256_checksums.csv |
SHA-256 checksums for all included files |
metadata/beacon_croissant.jsonld |
Croissant ML metadata record |
The status column of the manifest records how each file was processed: written (CSV and
JSON files rewritten with opaque IDs), typed_text_masked (keyboard CSVs), game_traffic_only
(PCAPs), and copied_unredacted (screen recordings).
Anonymization
- Participant and session folders use opaque sequential IDs
- CSV timestamps are raw Unix epoch values (not session-relative)
- Keyboard CSVs: the logger recorded every key, including text typed outside the game. Text
typed outside play (for example in log-in forms, browsers, and chat) is masked: each typed
character is replaced by
<letter>,<digit>, or<symbol>, in theKeyandConsecutive Keyscolumns. All rows and timings, special keys (key.space,key.enter, ...), and the keys used for play are kept. About 7% of character key presses are masked. - PCAP: only UDP packets to and from the game servers are kept; Ethernet/IP source and
destination addresses are replaced with deterministic pseudonyms in
10.0.0.0/8 - Hardware JSON: hostname, username, MAC, IP, HID, serial, UUID fields removed
- Screen recordings: not redacted (see Access)
- Raw consent files retained privately; see
consent_policy.md
Known issues
- Four screen recordings are truncated (recording stopped before the file index was written)
and do not play:
P003/S001,P003/S009,P006/S006, andP026/S006. P003/S014andP003/S015contain no gameplay: their recordings show only the game launcher and the desktop, and their PCAPs hold no game traffic.- Four PCAPs hold no usable game traffic.
P003/S014,P003/S015, andP009/S002contain no game packets. InP002/S001the game traffic ran inside a VPN tunnel (Cloudflare WARP, UDP 2408), which cannot be separated from other traffic and was removed with it.
Changelog
2.1.1 (November 2026)
- Masked text typed outside play in the keyboard CSVs (see Anonymization). If you downloaded keyboard CSVs before release 2.1.1, delete them and use this release.
- The repository history was rewritten, so earlier revisions are no longer available.
2.1.0 (November 2026)
- Removed
P015/S002, which was a copy ofP016/S001created during processing. The release now has 78 sessions. - Replaced 19 files that still held data from an earlier upload with a different ID mapping
(in
P001,P004,P007,P008,P011,P016,P018, andP019). Files downloaded before this release for these participants may belong to other participants. - PCAPs now keep only the game's UDP traffic; DNS and other traffic were removed.
- Enabled access requests; regenerated the manifests, checksums, and Croissant record.
2.0.0 (18 May 2026) — previous release.
Citation
If you use this dataset, please cite it:
Singh, I., Kaur, G., Atwal, U. P. S., Singh, G., Singh, G., & Singh, M. (2026).
BEACON: A Multimodal Dataset for Learning Behavioral Fingerprints from Gameplay Data
[Data set]. Zenodo. https://doi.org/10.5281/zenodo.20034625
Singh, I., Singh, G., Kaur, G., Atwal, U. P. S., Singh, G., & Singh, M. (2026).
BEACON-Logger: A Behavioral Authentication and Network Traffic Logger for Game Environments.
Zenodo. https://doi.org/10.5281/zenodo.20062628
Please cite the following paper also:
@misc{singh2026beaconmultimodaldatasetlearning,
title={BEACON: A Multimodal Dataset for Learning Behavioral Fingerprints from Gameplay Data},
author={Ishpuneet Singh and Gursmeep Kaur and Uday Pratap Singh Atwal and Guramrit Singh and Gurjot Singh and Maninder Singh},
year={2026},
eprint={2605.10867},
archivePrefix={arXiv},
primaryClass={cs.CR},
url={https://arxiv.org/abs/2605.10867},
}
License
Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
- Downloads last month
- 2,849