Five checks, and a block that fails one of them was never a block

The chain's tip lives in a single Cardano UTxOAn unspent transaction output, the unit Cardano's ledger is made of: an amount, the data attached to it, and the address that governs it., identified by a state-thread token and held at the validator script. The data attached to it records the tip hash, the height, the current base targetThe number a header's hash has to come in under for the block to be valid; a lower target means more work. and the Cardano time of the last block.

A block is accepted when a transaction spends that output with the header as its redeemer and the script confirms five things.

whitepaper §5

Five checks, in order

1 · Linkage
prev_hash equals the stored tip hash. A block extends the chain or it is not a block.
2 · Work
SHA-256d(header) is at or below the target the header committed to. This is the proof of work itself, recomputed on chain.
3 · The target was allowed
The committed target is no easier than the protocol allows at the start of this transaction's validity intervalThe window of chain time within which a Cardano transaction may be accepted, which is where WHAT's clock comes from., and that interval is no wider than [W = 120 s] The maximum interval width the validator accepts. A working value; see the note below. That is where the difficulty rule is enforced, against time Cardano attests to rather than a timestamp the miner reported.
4 · The commitment
The coinbase commitment carries the right chain tag and height, and its hash equals merkle_root. This is what puts the payout address and its datumData attached to a Cardano output; the WHAT tip's datum holds the tip hash, the height, the base target and the time of the last block. inside the work.
5 · Payment and the new tip
The transaction mints exactly the block reward, puts it in the one output that holds WHAT, at the committed address under the committed datum, and writes the new tip back to the script — not more, not less, and nowhere else.

All five are checked by Cardano when the transaction is validated. If any fails, the transaction is invalid and nothing is minted. What each one closes off is set out check by check in whitepaper §5.

They are also run a second time, on the last block, by the chain status page — on the header and coinbase bytes the chain carries, with the same protocol code, and each result shown separately. A check there does not make a block valid; Cardano already decided that. It is there so that a figure on the page is never the only evidence the block was read correctly.

Check 3 is worth being exact about, because the miner picks the number it reads. The interval's start is the miner's choice; what the validator does is cap how far the interval may stretch and then hold the target to its start. So "Cardano-attested" here means within [W] of a time Cardano enforced, not to the instant — and the enforcement is real, because stake-pool block producers reject a transaction outside its interval. See why choosing that start differently does not help a miner.

whitepaper §5

And why this chain has no forks of its own

Cardano supplies what a proof-of-work chain normally supplies for itself. Ordering and finality come from Cardano blocks. Time comes from the transaction's validity interval, which stake-pool block producers enforce.

Forks cannot occur. Two miners who solve the same height race to spend one output; the loser's transaction is simply invalid, which is the equivalent of a stale share. There is no competing history to choose between, so there is nothing for a fork-choice rule to do here.

One block per ten minutes is negligible load on Cardano. WHAT inherits Cardano's short rollbacks, so a block is final when its transaction is, and miner software must expect the tip to revert for a few Cardano blocks and resubmit. See how miners handle a reverted tip.

That is the whole mechanism. What is left is the list of things this project does not settle.