Entropy conditioning, pools and reseeding
Conditioning compresses raw entropy-source output so the result carries close to full entropy per bit, usually with a hash or block-cipher construction. Reseeding policy then decides how often that material refreshes the DRBG — an architecture decision that governs how long a state compromise stays exploitable.
Why raw output is never used directly
Raw output from a physical source is biased and correlated. Not because the source is poor, but because physical processes have structure: an ADC has offset, adjacent samples share circuitry, and the digitised signal contains classical noise alongside the process you are measuring.
Conditioning compresses that into a shorter, near-uniform output. The compression ratio comes from the min-entropy estimate: if raw samples carry a quarter of a bit each, four bits of raw output yield at most one bit of conditioned entropy, and a sound design leaves margin beyond that.
The trap in conditioning
A good conditioning function makes under-entropic input look excellent. Feed it half the entropy the design assumed and the output still passes every statistical test, because that is what hash functions do. The deficiency is invisible downstream, permanently.
This is the strongest argument for measuring the source rather than the output. After conditioning there is nothing left to measure — the evidence has been compressed away along with the structure.
Pools, and why they exist
An entropy pool accumulates material from one or more sources so that consumers can draw from it without blocking. Operating-system designs use several pools with different reseeding cadences, so that a fast consumer does not starve a slow but important one.
Pooling has a real benefit beyond buffering: mixing multiple sources means no single one being weak is fatal. It also has a cost — you can no longer say which source your key came from, which matters when you have to evidence provenance. That tension is the subject of key provenance.
Reseeding is a policy, not a default
How often a DRBG reseeds decides how long a state compromise remains exploitable. Reseed rarely and an attacker who recovers the state predicts output until the next reseed. Reseed constantly and you need entropy throughput to match.
Most systems inherit whatever their library chose. Treating it as a decision means knowing your consumption, which is arithmetic rather than intuition — and the arithmetic is worth doing once.
Keep reading
- Entropy fundamentals
How much entropy does cryptography actually need?
Entropy demand is calculable: DRBG instances multiplied by seed size divided by reseed interval, plus whatever consumes entropy directly, plus headroom for bursts. For most systems the answer is far below what a hardware source produces, which is exactly why sizing honestly is more useful than quoting a peak rate.
- Entropy fundamentals
What is an entropy source? Source, RNG and DRBG
An entropy source is the physical component that produces unpredictability. An RNG is the complete construction that turns that into usable output. A DRBG is the deterministic algorithm inside it that expands a seed. The three are routinely conflated, and untangling them makes the SP 800-90 series readable.
- Entropy fundamentals
Entropy vs randomness, a distinction that matters
Randomness describes how output looks; entropy measures how much unpredictability is actually present. A stream can be perfectly random-looking and contain no entropy at all — which is precisely what a good pseudorandom generator produces, and why statistical testing cannot establish security.