How to assess the maturity of an RNG
Assessing a random number generator means establishing four things: where unpredictability physically originates, what entropy estimate the vendor claims and what model it is conditional on, what evidence exists that the source still matches that model in service, and what the certification scope actually covers.
Deliberately vendor-neutral. Run it against us as readily as against anyone else — a checklist only one supplier can pass is a specification, not an assessment.
1 · Where does unpredictability originate
- Name the physical process, not the product. "Quantum" is not an answer; "spontaneous emission in a gain-switched laser" is.
- Ask what happens to the output if that process stops. If the system continues producing plausible output, ask what detects it.
- For a deterministic generator, ask what seeds it, how often it reseeds, and what happens at first boot.
2 · What is the entropy estimate conditional on
- Ask for min-entropy per raw sample, not per output bit. Output bits are post-conditioning and the number is not comparable.
- Ask what the estimate assumes. Every honest estimate is conditional on a model; a vendor who says theirs is not is telling you they have not written the model down.
- Ask how much of the digitised signal is classical noise and whether it is counted as entropy.
3 · What evidence exists at runtime
- Which health tests run continuously, and what failure modes do they actually catch? Repetition and adaptive-proportion tests catch a stopped source, not a degraded one.
- Are the physical quantities the entropy model depends on measured in service, or only at evaluation? This is the Level 3 to Level 4 line.
- Can the system report its own entropy health to your monitoring, or does it only fail silently or completely?
4 · What does the certificate cover
- Which document, which version, which laboratory, and which specific configuration was evaluated?
- Does the scope cover the entropy source, the DRBG, or the whole module? These are commonly conflated in marketing material.
- Was the evaluated configuration the one being sold to you?
Using the answers
A source can score modestly on this and still be the right choice, if the answers are honest and the gaps are ones your threat model tolerates. What should worry you is not a low score but an unanswerable question — particularly a vendor who cannot state the model their entropy estimate rests on. The buyer's guide turns this into twelve questions you can put in a document.
Keep reading
- The four levels of RNG
Level 3 vs Level 4, what observability adds
Level 3 means a physical entropy source was characterised during evaluation. Level 4 means the quantities its entropy model depends on are also measured while the system runs. The difference is whether your evidence describes the device on the day it was tested or the device that is generating your keys now.
- The four levels of RNG
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.
- The four levels of RNG
The 4 Levels of RNGs: How Random Is Your Random?
Quside develops new-generation quantum technologies for secure connectivity and efficient computation.