Skip to content

How to verify an entropy source you did not build

Verifying a source you did not build means asking for four things: the physical model, the min-entropy estimate and what it is conditional on, the runtime health evidence, and the exact scope of any certificate. For most sources on the market at least one of these has no good answer.

A vendor cannot be expected to volunteer the weakest part of their story. These are the questions that locate it.

Ask for the model, not the marketing

Which physical process, described specifically enough to be disagreed with. Which measurable quantities the entropy estimate depends on. What happens to the estimate if any of them moves.

A vendor unable to answer has not written a model down, which means their entropy figure came from measurement without a framework or from a datasheet they inherited.

Ask what the number is conditional on

Min-entropy per raw sample, the estimator that produced it, the conditions of measurement, and how much of the signal was attributed to classical noise. What honest reporting looks like is the standard to hold them to.

Ask what runs in service

Which health tests, at what cadence, catching which failure modes. Then ask the specific question: what detects a source producing two-thirds of its modelled entropy? If the answer is the SP 800-90B continuous tests, the honest answer is nothing does.

Ask what the certificate covers

  • Which scheme and document version, which laboratory, which report number.
  • Which component — entropy source, DRBG mechanism, or whole module? These are frequently conflated.
  • Was the evaluated configuration the one being sold to you?
  • If the phrase "Level 4 certified" appears, ask which body issues it. None does; the levels are an assurance model.

What a good answer looks like

Specific, conditional, and willing to name limits. A vendor who says "our estimate assumes X and we monitor Y to detect X ceasing to hold, and our certificate covers the entropy source only" has given you something to evaluate. One who says their randomness is certified and true has given you nothing to check.

Apply this to us as readily as to anyone else. What an ESV certificate actually proves is where our own scope is stated.