l-e-y/ton-dex-high-frequency

TON high-frequency DEX message windows

Sender-address windows containing at least 15 recognized native DEX swap messages in ten minutes.

Publisher: Feel the Block · Version: 2024-01-01_2026-10-01 · Rows: 1,677,514

Project website: feeltheblock.cc

Coverage: 2024-01-01 through 2026-09-30 (UTC). Granularity: 10 minutes × sender.

Files

• data.parquet: complete table for analysis and the dataset viewer.
• data.csv: complete CSV table.
• The CSV files together contain exactly the same rows as the Parquet table.
• Empty cells represent unavailable / undefined values, not zero.

Methodology

Fixed UTC ten-minute windows, left boundary included and right excluded. Count native rows with is_swap=true and non-empty sender and dex_router. Keep windows with at least 15 messages. dex_router contains protocol/version labels, whose distinct count is published.

Scope and limitations

No amount or execution-status filter. Counts are recognized messages, not all successful economic swaps. A high count does not prove bot activity. Router/opcode recognition is independent of the address list used for DEX flows.

Fields

Field	Meaning
bucket_start	UTC start of the ten-minute window.
address	Public sending address.
recognized_swap_messages_count	Recognized native swap messages in the window.
dex_protocol_versions_count	Distinct recognized protocol/version labels, not router-address count.

Provenance and version

Derived by Feel the Block from locally preserved TON native message records and existing historical analysis outputs. Each file has a SHA-256 checksum in manifest.json. The snapshot is historical and immutable; UTC boundaries and denominators are defined above. This release is prepared for Hugging Face.

License and attribution

Public release is under CC BY 4.0 for the publisher's rights in the derived dataset. Upstream and third-party rights are not relicensed. See LICENSE-DATA.md for the publication condition and attribution.

Suggested citation: Feel the Block, “TON high-frequency DEX message windows”, release 2024-01-01_2026-10-01, dataset version and source URL.