? BlockDAG

How a blockDAG actually works

A blockDAG is the generalisation of a blockchain to a Directed Acyclic Graph: instead of every block having exactly one parent, each new block can reference an arbitrary set of previously-seen tips. Two miners that produce a block at roughly the same height both contribute to the structure rather than one of them being orphaned. The challenge is that DAGs admit multiple valid topological orderings, so a blockDAG protocol must specify a deterministic rule that turns the partial order into the single linear order that user-visible state transitions need. Every blockDAG design - Kaspa's GHOSTDAG, IOTA's Tangle, Sui's Mysticeti, Aleph Zero's AlephBFT - is essentially a different ordering rule layered on the same DAG primitive.

The canonical permissionless example is GHOSTDAG, the Phantom-family protocol that powers Kaspa. Each block points to every tip it has observed; the protocol scores a candidate 'selected tip' by total accumulated work and walks back from that tip identifying a 'blue set' - the largest k-cluster of mutually-referencing blocks where k is a network-tuned anti-cone parameter (Kaspa runs at k=18 today). Blocks in the blue set are accepted and ordered topologically; blocks outside it are 'red', remain in the DAG for fork-choice but earn no coinbase reward. The result is a Nakamoto-style proof-of-work chain that produces one block per second without an orphan tax - the parallel work is captured by the structure rather than thrown away - which is why Kaspa's effective throughput sits an order of magnitude above Bitcoin's despite using comparable per-block work.

The permissioned BFT variant takes a different shape. Sui's Mysticeti separates the mempool from consensus: validators produce a DAG of certificates (each a quorum-signed batch of transactions) round-by-round, and a lightweight commit protocol on top of the DAG selects a deterministic linear order of those certificates. Owned-object transactions skip the consensus DAG entirely and finalise via a Byzantine quorum signature on a single object's history, which is the trick that gives Sui its sub-second finality for the simple case. Aptos's Quorum Store and Aleph Zero's AlephBFT are variations on the same theme - DAG ordering as a throughput primitive that lets a permissioned validator set commit thousands of transactions per round without the leader-bottleneck of single-leader BFT.

Key concepts

DAG ordering rule

The function that turns the partial order of a DAG into a single linear order that consumers can read state from. GHOSTDAG uses a heaviest-blue-sub-DAG rule with anti-cone parameter k. Mysticeti uses a deterministic commit protocol on the validator DAG. The Tangle uses leaderless On-Tangle Voting. Without a deterministic ordering rule, two honest nodes could see the same DAG and disagree about which transaction came first, breaking the single-state model that smart contracts depend on.

GHOSTDAG and the blue set

GHOSTDAG classifies each block as blue (in the largest k-cluster of mutually-referencing blocks reachable from the selected tip) or red. Blue blocks earn coinbase, red blocks do not, but red blocks still contribute to fork-choice weight. The k parameter controls how 'wide' the DAG is allowed to be - Kaspa runs at k=18 with one block per second, meaning up to ~18 parallel blocks per second can all be blue. Higher k raises throughput but increases sensitivity to network propagation delay; lower k collapses toward a linear chain.

Tangle and leaderless ordering

IOTA's Tangle is a DAG of transactions (not blocks) where each new transaction must approve two prior tips. The original Tangle relied on the centralised Coordinator for finality; the Coordicide redesign (live as IOTA 2.0 / Stardust) replaces the Coordinator with leaderless On-Tangle Voting and a reputation-weighted committee. The Tangle is the most architecturally distinct mainstream DAG - it is permissionless without proof-of-work and has no notion of 'block' at all - and serves as the reference point for the broader leaderless DAG research line.

Mysticeti and Sui's causal-history ordering

Sui's consensus is a DAG-based BFT protocol. Validators run a certificate DAG (each round produces a block of certified transactions referencing prior-round blocks) and a commit protocol selects a deterministic order of vertices. The trick that gives Sui sub-second finality is that owned-object transactions (single-writer, no shared state) skip the DAG entirely and commit via a Byzantine quorum signature on the object's causal history. Shared-object transactions go through the full DAG-based consensus path. The split is the design choice that lets Sui claim sub-second finality without sacrificing safety on contested writes.

BDAG token and the BlockDAG project's staking

BlockDAG (BDAG) is a Layer-1 project with a 150-billion total supply, 25 billion reserved for staking rewards, mainnet live in 2026, and listings on 15+ exchanges. Staking flows: connect wallet -> select a pool -> commit BDAG -> earn epoch-based rewards -> claim after a four-epoch cooldown. As with any high-supply, freshly-listed staking token, the relevant numbers for evaluating yield are circulating supply at stake-time, annualised emission rate, and percentage of supply staked.

Throughput trade-offs and finality

The DAG primitive raises throughput by capturing parallel work that a linear chain would orphan. The cost is propagation: to safely raise the parallelism parameter (k in GHOSTDAG, the round-rate in Mysticeti) the network must disseminate blocks fast enough that honest miners' tips converge. High-k DAGs are therefore sensitive to network heterogeneity, and the practical operating point is a balance between throughput and the slowest-honest-miner's propagation latency. Finality semantics also differ: GHOSTDAG offers probabilistic finality like Bitcoin; Mysticeti offers single-slot deterministic finality once the commit protocol selects a vertex.

Why blockDAG designs matter

As of April 2026, blockDAG designs have moved from a research curiosity to the dominant scaling primitive for new Layer 1s. Kaspa, the flagship permissionless-PoW blockDAG, has held a top-30 market-cap slot since 2024 by demonstrating that a Nakamoto-style chain can sustain one-second blocks without giving up proof-of-work security. Sui and Aptos, the permissioned-BFT camp, have run the Mysticeti and Quorum-Store variants in production through enough volume to give the design line credibility against single-leader competitors like classical Tendermint. The IOTA Tangle remains the reference point for leaderless permissionless DAGs even though IOTA's consumer adoption has been slower than expected. Together they represent the three axes - permissionless versus permissioned, PoW versus BFT, leader-based versus leaderless - that any new L1 blockDAG design picks coordinates on.

For DeFi specifically, the practical consequence is faster finality and cheaper writes - a blockDAG L1 with sub-second finality lets perps and order books operate without the L2 round-trip Ethereum's L2 stack imposes. The architectural cost is that smart-contract platforms on blockDAGs (Sui Move, Aptos Move) reason about transactions in terms of object ownership and causal history rather than the global-state, single-thread model that Solidity assumes. Porting a Solidity protocol to a Move-on-DAG L1 is therefore not a straight translation; it is a redesign around object-oriented state. The tradeoff is real throughput at the cost of the EVM lock-in that has carried Ethereum's DeFi network effects since 2020, and how fast the DeFi ecosystem migrates to DAG-based L1s remains the open question of 2026.

Frequently asked questions

What is a blockDAG and how is it different from a linear blockchain?

A blockDAG (block Directed Acyclic Graph) is a consensus structure where each new block can reference more than one parent. The result is a partial order over blocks instead of the strict linear chain that Bitcoin and Ethereum produce. Two miners who solve a block at roughly the same time both contribute their work to the structure rather than one of them being orphaned. The trade-off is that a blockDAG has to define a deterministic rule for ordering parallel blocks before users can read a single 'state', which is exactly what protocols like GHOSTDAG (Kaspa), the Tangle (IOTA), and the causal-history scheme (Sui) each solve in their own way.

How does Kaspa's GHOSTDAG protocol pick a final ordering of parallel blocks?

GHOSTDAG (Greedy Heaviest-Observed Sub-DAG) is the Phantom-family protocol that powers Kaspa. Each new block points to all 'tips' it has observed, and the protocol classifies blocks as 'blue' (in the largest k-cluster of mutually-referencing blocks) or 'red' (outside it). The blue set is selected greedily by walking back from the highest-scoring tip, and within the blue set blocks are ordered topologically. Kaspa's mainnet runs at one block per second with k=18, meaning up to roughly 18 parallel blocks per second can all be 'blue' and contribute coinbase rewards instead of being orphaned. The protocol's selling point is that it preserves Nakamoto-style proof-of-work security while eliminating the orphan tax, and its design influenced the broader 'high-throughput PoW' research line.

How does IOTA's Tangle differ from Kaspa's GHOSTDAG?

IOTA's Tangle is a DAG of individual transactions (not blocks), where each new transaction must reference and validate two prior 'tips' before it is broadcast. Originally the Tangle was leaderless - no miners or validators - but the design suffered from a need for the Coordinator, a centralised reference node, to provide finality. The 2.0 redesign (Coordicide, shipped as IOTA 2.0 / Stardust) replaces the Coordinator with a leaderless committee that runs On-Tangle Voting and a reality-based ledger. Compared with GHOSTDAG, the Tangle is permissionless without proof-of-work, and ordering is established by reputation-weighted voting on conflicting transactions rather than by a heaviest-sub-DAG rule on blocks.

Is Sui's consensus a blockDAG and how does it order parallel blocks?

Sui's consensus is a DAG-based BFT protocol - Mysticeti as of the 2024 upgrade, succeeding Narwhal+Bullshark. Validators run a mempool DAG (Narwhal) where each round produces a 'block' of certified transactions referencing the prior round's blocks, and a separate consensus protocol (Bullshark, then Mysticeti) deterministically commits a sequence of those DAG vertices as the canonical order. Owned-object transactions (single-writer) skip consensus entirely and finalise via a Byzantine-quorum signature, which is why Sui claims sub-second finality for the common case. The DAG here is a throughput optimisation on a permissioned BFT validator set, not a permissionless miner DAG like Kaspa's.

How does the BlockDAG (BDAG) project's tokenomics work?

The BDAG token has a 150-billion total supply with 25 billion (about 16.7%) reserved for the staking-rewards pool. The mainnet went live in 2026, and the project lists on more than 15 exchanges. Stakers commit BDAG to a validator pool, earn epoch-based rewards from the rewards pool, and can withdraw after a four-epoch cooldown. As with any high-supply, exchange-listed staking token, the relevant numbers for evaluating yield are the circulating supply at the time you stake, the annualised emission rate from the rewards pool, and the percentage of stakers - high-supply schedules typically dilute nominal APY meaningfully in the first 12-24 months.

What is the practical throughput advantage of a blockDAG over a linear chain?

On a strict linear chain, throughput is bounded by block size divided by block time: any work done in parallel by two miners is wasted because one block is orphaned. On a blockDAG with k=18 like Kaspa, up to ~18 parallel blocks per second can all contribute, raising the effective throughput by roughly k without changing the per-node verification work for any single block. The catch is propagation: to safely raise k you need fast block dissemination across the network, which is why high-k blockDAGs are sensitive to network heterogeneity. Sui's Mysticeti reports a similar idea applied to a permissioned BFT setting: parallel certificate production decouples throughput from the latency of a single round.

As of April 2026, where do blockDAG designs sit relative to mainstream L1s?

As of April 2026, blockDAG designs occupy two distinct niches. Permissionless PoW blockDAGs - Kaspa is the canonical example, with a market cap that has put it inside the top 30 by capitalisation since 2024 - compete on raw throughput against Bitcoin's linear chain. DAG-based BFT designs - Sui's Mysticeti, Aptos's Block-STM-on-Quorum-Store, Aleo's snarkOS - sit inside permissioned validator sets and compete with Solana and Ethereum L2s on user-experience metrics like sub-second finality. The shared lesson is that DAG ordering is now considered the default scaling primitive for new L1s; the design space has fragmented into permissionless-PoW, permissioned-BFT, and reputation-based variants rather than collapsing into a single 'blockDAG' winner.

Related reading

• BlockDAG architecture - DAG vs linear chain, ordering rules, and how parallel blocks become a single canonical state.
• BDAG staking guide - step-by-step pool selection, epoch rewards, and the four-epoch claim cooldown.
• Layer-2 DeFi - how rollups and DAG-based L1s compare on finality, throughput, and developer experience.
• Perpetual DEX primitives - why sub-second finality changes what perp DEXs can do at the matching layer.
• Oracle internals - push vs pull oracles and why Pyth-style pull designs pair naturally with high-throughput DAG L1s.