No one is chosen in advance. Every online account runs a private lottery each round; a handful win the right to propose, and two sampled committees vote a block final in under three seconds.
Click a step on the right, or use the arrow keys, to jump to it. Move your mouse to shift the view.
Counts the top-level transactions in each block by type: payments in ALGO, asset transfers (ASAs), application calls, key registrations, asset configuration and freezes, state proofs and heartbeats. Inner transactions sent by smart contracts are not counted separately.
| Round | Transactions per type |
|---|
Use the left and right arrow keys to read each round's count.
| Round | Transactions | Application calls | Other | Proposer | Time |
|---|
Any account can take part in consensus by registering a participation key and going online. There is no minimum to participate and no delegation to a fixed set of validators; influence is simply proportional to the ALGO an account holds online. Around 1,400 accounts are online with roughly 2 billion ALGO between them, a fifth of the supply. Because nobody is chosen ahead of time, there is no known target to attack or bribe before a round begins.
A verifiable random function takes a private key and an input (here, the previous block's seed and the round number) and returns a random-looking output plus a proof. Anyone with the public key can check the proof, but nobody can predict or forge the output. Each account compares its output against a threshold scaled by its stake: a small holder rarely wins, a large holder wins often, and every unit of ALGO has the same chance. The protocol targets about 20 proposers per round; the one with the lowest credential leads.
Byzantine Agreement runs in steps, and every step draws a new committee by the same lottery. The soft-vote committee has an expected 2,990 votes and needs 2,267 to agree on the lowest-credential proposal; the certify committee has an expected 1,500 votes and needs 1,112 to make the block final. A vote counts as many times as the account won the lottery, so stake, not node count, decides. If a step times out without a quorum, the round moves to a new period and larger recovery committees take over until agreement resumes.
A block exists only once it is certified, and certification needs more than two thirds of a stake-weighted committee to vote for exactly one proposal. Two conflicting blocks can't both reach that bar, so the chain never forks and nothing is ever reorganised. Rounds take about 2.8 seconds on average since the dynamic round-times upgrade in 2024, and a transaction is settled the moment its block appears. Each block carries a seed that the proposer derives with its VRF from the previous seed, and the next round's lottery uses it, which is what stops anyone steering the draw.
Participation nodes hold the keys and do the voting. Relay nodes hold no stake; they are well-connected hubs that collect transactions, proposals and votes and forward them to every node connected to them, so a message reaches the whole network in a couple of hops. Anyone can run either kind. Relays only route messages, so a bad relay can delay the network but can't sign anything on its behalf.
Since the consensus upgrade of January 2025 (go-algorand 4.0), the protocol pays block proposers directly. An account is eligible if it registered with a 2 ALGO fee and holds between 30,000 and 70 million ALGO online. A winning proposer receives half the fees in its block plus a bonus that started at 10 ALGO and decays 1 % every million rounds (about 8.3 ALGO now), paid from the fee sink. Fees are not burned: they go to the fee sink, which funds the payouts. Accounts whose nodes stop proposing are marked absent and suspended; nodes prove liveness with heartbeat transactions when they haven't proposed for a while. The minimum fee is 0.001 ALGO, and since the 5.0 upgrade of August 2026 large transactions pay a surcharge per byte.
Live mode polls a public Algorand node over HTTPS about once a second, fetches every new block, and backfills the last 40. The round numbers, timestamps, transaction counts and types, transaction IDs, fees, proposer addresses and proposer payouts are real, and so is the online stake. Votes travel privately between participation nodes and aren't served by the API, so the sortition sparks, the committee beams and the threshold arcs are re-enacted for each real round. The ring shows a sample of 40 of the roughly 1,400 online accounts; the node that lights amber is the account that really proposed the block, placed on the ring by its address.
Real committees are chosen per vote rather than per node, and a large account can hold many votes at once; the beams here light one node per vote. Steps overlap in practice: soft votes for one proposal can start before every proposal has arrived, and the propose, soft and certify steps each have their own timeouts. Recovery periods, state proofs and the exact fee-sink accounting are left out. Read the Algorand consensus documentation for the full protocol.