Skip to content

What is entropy assurance?

Entropy assurance is the combination of three things: a physical model of where unpredictability originates, measurement of the quantities that model depends on while the system runs, and evidence a third party can inspect. Any two without the third leaves a gap somebody will eventually ask about.

The industry has no settled vocabulary for the difference between a source that was tested once and a source you can trust right now. This is the term for it.

One · a physical model

A statement of which physical process produces unpredictability, and what quantities the resulting entropy estimate depends on. Not "quantum" — a description specific enough that someone could disagree with it.

A model is what makes an entropy estimate meaningful. Without one, a min-entropy figure is a number with no argument attached, and there is no way to say what would make it wrong.

Two · measurement in service

The quantities the model depends on, measured while the system runs. Not the statistics of the output — those survive severe degradation unchanged — but the physical parameters the entropy claim rests on.

This is the component almost universally missing. Most sources are characterised during evaluation and then assumed. The distinction between characterised and observed is the whole of it.

Three · inspectable evidence

Something an auditor, a customer or a certification body can examine: validation reports, health-test records, telemetry retained long enough to answer a question about a key generated last quarter.

Evidence is what turns the first two into something usable in procurement. A vendor who has a model and measures it but retains nothing cannot answer the question that matters after an incident.

Why all three

Model plus evidence without measurement is a certificate describing the past. Measurement plus evidence without a model is telemetry nobody can interpret. Model plus measurement without evidence is a good engineering story that fails an audit.

Assembled, they answer the question that no amount of statistical testing can: not "does this look random" but "what do you know about this source, how do you know it, and can you show me". Locally Verified Entropy is the product-shaped version of that answer.