Skip to content

The four levels of RNG, a maturity model for randomness

The four levels describe how much you can establish about a random number generator, from statistical output testing at Level 1 to a source with a physical entropy model that is monitored while it runs at Level 4. It is a maturity model for reasoning about assurance, not a certification scheme.

Nobody issues a "Level 4 certificate". The levels are a way of asking what evidence exists about a generator, and they map onto the language that NIST and BSI already use rather than replacing it.

Level 1 — output testing only

The generator produces a stream that passes statistical test suites. That is all that is established. A deterministic generator with a known seed sits comfortably at this level, which is precisely why the level is weak: it says something about the output and nothing about unpredictability.

Level 2 — a deterministic construction on a trusted seed

A specified DRBG mechanism, correctly implemented, seeded from a source that is asserted to be adequate. This is where a great deal of well-built software sits, and it is a reasonable place to be. The gap is that the seed's quality is assumed rather than characterised.

Level 3 — a characterised physical source

The entropy source is physical, and it has been analysed: there is a stated model, an entropy estimate derived from it, and the device was tested against that model during evaluation. This is the level a conventional SP 800-90B validation or an AIS 31 evaluation establishes.

The limitation is when the evidence was produced. Characterisation happens at evaluation time on sample devices. It says what the source did then, under those conditions.

Level 4 — an observable source

Everything at Level 3, plus the physical quantities the entropy model depends on are measured while the system runs, so a departure from the model can be detected in service rather than inferred from a certificate issued earlier. What that actually adds is the distinction most worth understanding in this model.

How to use the model

Grade any source, including ours, by asking what evidence exists and when it was produced. The assessment checklist turns that into something you can run in a procurement process.