README.md

metadata

license: cc-by-4.0
task_categories:
  - tabular-classification
  - time-series-forecasting
language:
  - en
tags:
  - tsfile
  - timeseries
  - internet-censorship
  - network-measurement
  - internet-freedom
  - ooni
  - ioda
  - censoredplanet
  - time-series
  - format:tsfile
pretty_name: Voidly Benchmark v1 (TsFile)
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: evidence
        path: evidence.tsfile
      - split: incident_evidence
        path: incident_evidence.tsfile
      - split: incidents
        path: incidents.tsfile

Voidly Benchmark v1 (TsFile)

Apache TsFile version of emperor-mew/voidly-bench-v1.

Overview

An evidence-backed internet-measurement benchmark: individual observations from public measurement sources (IODA, OONI, …) with their signal level and confidence, the incident-to-evidence links that group them, and the resulting incident records with severity, mechanism and duration.

• Tables: 3 TsFile table(s), 86,546 rows in total.
  • evidence: 49,596 rows, 2016-07-15 00:00:00 → 2026-04-16 18:00:00.093000 (UTC)
  • incident_evidence: 35,376 rows, 1970-01-01 00:00:00 → 2026-04-17 03:47:14.498000 (UTC)
  • incidents: 1,574 rows, 2016-07-15 00:00:00 → 2026-04-12 08:00:00.001000 (UTC)

Schema (TsFile structure)

evidence.tsfile

• Time (INT64, milliseconds) — the source observed_at column.
• TAG: none — the table is stored as a single device.
• FIELD (21): id (int64), source (string), kind (string), source_ref (string), source_url (string), retrieved_at (string), country_code (string), region_code (string), asn (double), signal_type (string), signal_value (double), signal_baseline (double), signal_level (string), upstream_severity (string), domain (string), continent (string), un_subregion (string), risk_tier (double), domain_category (string), blocking_method (string), confidence (double)

incident_evidence.tsfile

• Time (INT64, milliseconds) — the source linked_at column.
• TAG: none — the table is stored as a single device.
• FIELD (4): incident_id (string), evidence_id (int64), link_confidence (double), link_reason (string)

incidents.tsfile

• Time (INT64, milliseconds) — the source first_seen column.
• TAG: none — the table is stored as a single device.
• FIELD (18): id (string), readable_id (string), country_code (string), title (string), incident_type (string), mechanism (string), severity (string), confidence (double), status (string), last_seen (string), duration_hours (double), affected_domains (string), affected_services (string), affected_asns (string), measurement_count (int64), anomaly_rate (double), sources (string), revision (int64)

Conversion notes

• One TsFile per source Parquet (3 tables), each stored as a single device.
• Time = the observation timestamp of each table as INT64 milliseconds: observed_at (internet-measurement evidence), linked_at (incident-to-evidence links) and first_seen (incidents). incidents.last_seen is kept as a FIELD so both ends of an incident survive.
• No column is dropped apart from affected_services / affected_asns, which the source stores with Arrow type null and which are empty in every row.
• Not converted: reproduce.py, reproduce_results.json, source_reliability.json, DATA_HASHES.txt, LICENSE — the reproduction script, its recorded results and integrity hashes are not time series; they are mirrored into the repository as reference files.

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("evidence.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: emperor-mew/voidly-bench-v1
• Source card title / publisher: Voidly Benchmark v1
• License: cc-by-4.0 (as declared by the original dataset).