Open data

Everything behind the numbers, free to download.

Seven rolling tables as CSV under Creative Commons Attribution 4.0. Every row carries its own timestamps, so a download is a citable snapshot. Use them, check our work, publish your own analysis.

Format: CSV, UTF-8
Licence: CC BY 4.0
Refresh: 10 min to hourly

Services

One row per one monitored service

CSV

service_id: Stable id
name, category, probe_url: What is probed
organisation: Registrable domain of the operator
hosting_provider, hosting_region, hosting_country: Inferred from IP ownership and published cloud ranges
cdn, dns_provider, email_provider: Inferred from DNS
mapped_at: Last infrastructure mapping

Official incidents

One row per one incident from a company status page

CSV

organisation, status_page, incident_id: Who declared it and where
title, impact, status: As published
created_at, started_at, resolved_at: UTC timestamps
components, link: Affected components; link to the incident

Provider incidents

One row per one incident published by an infrastructure provider

CSV

source, incident_id: AWS Health, Google Cloud, Cloudflare, Akamai, Vercel
provider, region: Normalised provider name; region if published
title, severe: As published; severe = significant / major or worse
started_at, declared_at, resolved_at, link: UTC timestamps and source link

Upstream signals

One row per one live signal (rule cs-v1)

CSV

signal_id, rule: Stable id and rule version
opened_at, fired_at, last_at: First declaration, moment the rule fired, last joining declaration
likely_provider, layer, companies: Top candidate and the companies involved
p_value, evidence, minutes_vs_provider: Chance probability, evidence tier, timing vs provider

Radar sweeps

One row per one five-minute sweep

CSV · 31 days

run_id, at: Sweep id and time
probes, failed: Counts
vantage_suspect: More than 50% failed: excluded from analysis
failing_service_ids: Semicolon-separated

Outage episodes

One row per two or more consecutive failed sweeps of a service

CSV

service_id: Service
started_at, last_failed_at, recovered_at: UTC timestamps
failed_sweeps, reason: Length and most common error

Hourly reachability

One row per one service-hour

CSV · 7 days

service_id, hour: Service and UTC hour
probes, reachable: Counts
p50_ms, p95_ms: Latency of successful probes

Machine access

No key needed. Please cache responses; tables change at most every ten minutes.

What	Request
Any table as CSV	`GET /_api/research/dataset?table=<name>[&days=1..31]`
Every research number (JSON)	`GET /_api/research/overview`
Upstream signals and track record (JSON)	`GET /_api/upstream/signals`
Live radar status (JSON)	`GET /_api/radar/status`

Cite this work

ShadowGraph (2026). Do companies on the same infrastructure fail together? Shared fate and upstream signals in official incident records. ShadowGraph Research, rules sf-v1, cs-v1, val-v1, conc-v1. https://shadowgraph-live.floot.app/research

@techreport{shadowgraph2026research,
  title  = {Do companies on the same infrastructure fail together?},
  author = {{ShadowGraph}},
  year   = {2026},
  note   = {Rules sf-v1, cs-v1, val-v1, conc-v1; data CC BY 4.0},
  url    = {https://shadowgraph-live.floot.app/research}
}

Official incident titles are quoted from the companies' and providers' public status pages and remain theirs; every row links to its source.