Two equations, and no retarget

WHAT targets one block every ten minutes. Difficulty adjusts on every block, in both directions, by a single exponential rule in log space, evaluated against Cardano-attested time. There is no epoch and no retarget.

The chain stores a base targetThe number a header's hash has to come in under for the block to be valid; a lower target means more work. T and the Cardano time of the last block. With e the seconds elapsed since that time, Δ the interval the block being submitted closes, and H the half-life:

§6, the whole rule

Tallowed(e) = T · 2max(0, e − 600) / H
T ← T · 2(Δ − 600) / H

whitepaper §6

One line a miner evaluates, one the chain applies

The first line is what a miner works against between blocks: a closed-form function of the clock that every miner evaluates locally. No difficulty state changes while it runs and no Cardano transaction occurs. Past the ten-minute mark the allowed target loosens by one doubling per half-life, continuously.

The second line is the only state change, applied on confirmation from the on-chain timestamps alone. For a late block the two agree exactly — the update lands where the in-flight curve had already walked to — so the rule is continuous through a slow block. For an early block the in-flight term is inert and the update tightens the base by the mirror of the decay a late block would have earned.

That symmetry is what keeps the rule unbiased: it stops pushing in either direction only when the average interval is the ten minutes it aims at. Simulation over 400 blocks holds a 601-second mean at H = 10 minutes, with medians near 0.69 of the mean, which is where an exponential distribution puts them.

One useful thing follows from the first line. The allowed target never tightens as the clock runs, so the target when a block confirms is never harder than it was when the miner found it — a hash found now still validates when it lands in the next Cardano block, twenty seconds or so later.

Which raises the obvious question, since e is read from a validity intervalThe window of chain time within which a Cardano transaction may be accepted, which is where WHAT's clock comes from. whose start the miner chooses: why can a miner not simply claim a different one and help itself to the decay? Moving the start earlier only tightens the target the miner has to have met, so it is a straight loss. Moving it later does buy a looser target — but a transaction cannot be submitted before its interval opens, so the miner waits for Cardano's clock to reach it, which is the same as not having moved it. That is why the validator does not have to police where the interval sits, only how wide it is. See the check that enforces it.

Fig. 3 — The allowed target against the clock. The curve is line 1 as the seconds tick by. Drop a block in at an interval Δ and you see where line 2 lands: exactly on the curve for a late block, and below the starting level for an early one. H is adjustable across the swept 5-minute-to-1-hour range and defaults to the chain's 10 minutes. The vertical axis is the target as a difficulty ratio. This is the rule being evaluated, not a record of anything that happened.

whitepaper §6

Why H is ten minutes

Because ten minutes is one target interval, which is Bitcoin's block time. The value is inherited, not derived: Bitcoin has no half-life of its own to copy, and H and the target spacing are one constant in the validator rather than two.

Simulation swept five minutes to one hour. The mean holds at 601 to 608 seconds throughout, so the block rate does not turn on the choice. What the choice moves is the tail, and the tail tightens as H falls — seven intervals past three times the target at five minutes, against sixteen at an hour — at the cost of moving difficulty further on every block. So the sweep favours the short end and does not pick ten minutes.

H is compiled into the validator beside the ten-minute target, so it is part of the chain's identity: a different H is a different chain, not a setting anybody turns later.

That is the rule. The two questions people actually arrive with are what happens when a lot of hashpower turns up at once, and whether it pays to stay away and come back later.