Styxis
Company
Knowledge

Cloud availability still needs a separate business-data continuity check

Observe service availability and business-data arrival, completeness and timeliness separately during an independence exercise.

What a successful connection does not answer

A reachable service and complete, timely business data are different observations. Keeping them separate makes an independence exercise more useful to operating teams. Truthound can check a prepared arrival log and expected population, but cannot certify infrastructure independence.

The exercise AWS describes

AWS reported on November 1, 2026 that its October 24, 2026 European Sovereign Cloud exercise disconnected two global systems while workloads and customer data access continued. This is the supplier's account of its tested scope. It should not be expanded into separate proof that every customer's business dataset was complete.

List what the business actually awaits

For a local continuity test, define which data should arrive first. Separate batch, collection path, observation interval, original production time and arrival time. A set of probes returning success may differ from the datasets a business process actually consumes.

Keep the observation boundary stable

Use the same definitions before, during and after the exercise. Record a changed collection frequency or smaller observed population alongside the result. Do not add normally absent data to manufacture failures, or remove necessary data because it makes the test look worse.

The expected set in Truthound

Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides DistinctValuesEqualSetValidator for a supplied expected category set and CompletenessRatioValidator for a minimum non-null ratio. Operators define which batches were due and acceptable missingness. Neither check determines whether a network path is sovereign or independent.

Keeping the existing predicate

An existing Pandera 0.35.0 pandas predicate can keep checking arrival logs. Compare both paths with the same expected batches and time boundaries, including late arrivals, absent probes and repeated records. Support for custom checks does not establish that the same continuity contract has been implemented.

Late is different from permanently missing

Late arrival is different from permanent absence. Separate completeness now from completeness at the operational cutoff, so teams can distinguish reprocessable work from work already affected. Overwriting an earlier observation with a later arrival would hide the delay.

Report two dimensions

Report service availability and data continuity as two dimensions. A pass on one should not conceal an evidence gap on the other. Leave unobserved scope unconfirmed and identify which question belongs to infrastructure owners and which to business-data owners.

Specify the next experiment

The next experiment can refine which delays are tolerable and which missing inputs should stop a process. A bounded observation history makes everyday operating choices easier to explain. One successful connection should not become a substitute for every kind of continuity evidence.

Conditions

Define permitted observations, expected data and time boundaries for the organisation's test.

Alternatives

Retain existing Pandera 0.35.0 continuity predicates and compare them against the same expected population.

Limitations

  • Checks do not certify cloud sovereignty or failover; the AWS account is the supplier's report of its exercise.

Evidence

Event checked
Subject checked
Modified