Do companies on the same infrastructure fail together?
Dependency maps only matter if shared infrastructure becomes shared failure. We test that with — official incident declarations by — companies over — days.
Loading the latest results…
Same CDN—p —
Same DNS provider—p —
Same cloud provider—p —
Same cloud region—p —
How the test works
Three ideas, each chosen so the test cannot flatter us.
- 1Count close pairs. For every group of companies that share a provider, count pairs of incidents from two different companies declared within — minutes of each other.
- 2Build a fair “chance” world. Shift each company's whole incident history by a random number of whole weeks. Every company keeps its own incident rate and the working-hours rhythm; only the alignment between companies is destroyed. Repeat 1,000 times.
- 3Compare. Lift = real pairs ÷ average pairs in the chance worlds. The p-value is the share of chance worlds with at least as many pairs as reality.
Results by layer
Shaded rows are the controls. Random groups of the same sizes must show no excess; four synthetic shared events planted across ten random companies must be found.
Unit: one company. Only status pages whose public history covers the full window are used (— pages); pages mirrored for one company count once. Impact minor or worse.
Provider by provider
Every provider with at least five companies in the sample, tested separately. With many tests some will look significant by luck, so p-values are corrected (Holm–Bonferroni).
In the companies' own words
Incident titles that name the same upstream provider within six hours on at least two independent status pages. — such events in the window.