Consensus
DAGKnight, explained
Kaspa's next consensus protocol, in plain English: what it is, how it beats GHOSTDAG, and when it's coming.
Status In development. Kaspa runs on GHOSTDAG today. DAGKnight is being built in the Rust node and has no mainnet date yet. Updated 2 November 2026.
DAGKnight in one minute
Kaspa produces blocks so fast (10 every second) that many are mined at the same moment. Instead of throwing the extras away like Bitcoin would, Kaspa keeps them all in a blockDAG and uses a consensus protocol to put them in order.
Today that protocol is GHOSTDAG. It works well, but it relies on a number, k, fixed in advance from a guess about the worst network delay. DAGKnight removes that guess. It reads the network's real behaviour from the DAG and adapts, so confirmations are fast when the network is healthy and stay safe when it isn't.
GHOSTDAG vs DAGKnight
| GHOSTDAG (today) | DAGKnight (planned) | |
|---|---|---|
| Key setting | Fixed parameter k, set from an assumed worst-case delay | None. Parameterless |
| Network delay | Must stay within the assumed bound to remain secure | Adapts to whatever the real delay is |
| Confirmation speed | Paced by the worst-case assumption | Paced by actual network conditions |
| Attacker threshold | Below 50% of hashrate, while delays stay within the bound | Below 50% of hashrate, no fixed latency bound |
| Proof of work | kHeavyHash | kHeavyHash (unchanged) |
| Published | 2018 paper | 2022 paper |
Why a fixed parameter is a problem
Think of k as a speed limit chosen before anyone has seen the road. Set it too low and honest blocks mined in parallel during a slow patch get treated as suspicious, which weakens security. Set it high enough to cover every bad day and the network has to wait longer than necessary on every good day.
GHOSTDAG picks a safe, conservative k. That costs confirmation speed most of the time, and if real delays ever exceed the assumption, the security guarantees no longer fully hold.
How DAGKnight works, in plain English
- Look at the DAG. Honest miners see each other's blocks quickly, so their blocks form a tightly connected cluster. An attacker mining in secret produces blocks that are poorly connected to the rest.
- Search for the right k. Instead of a fixed k, DAGKnight looks for the smallest k at which a group of blocks backed by a majority of the work fits together as a well-connected cluster. That k reflects the network's real conditions right now.
- Order the blocks. With that k found, blocks are ordered much like GHOSTDAG would order them, favouring the well-connected majority cluster.
- Break ties carefully. When competing clusters come out even, a tie-breaking rule decides. This is one of the parts currently in code review.
Where DAGKnight is up to
dagknight branch of rusty-kaspa, including the tie-breaking rule.Frequently asked questions
Is DAGKnight live on Kaspa mainnet?
Not yet. As of 2 November 2026, DAGKnight is being implemented on a dedicated branch of the Rust node, with parts of the protocol still in code review. No mainnet activation date has been announced. Kaspa currently runs on GHOSTDAG. Check our upgrade tracker for the latest.
Who created DAGKnight?
DAGKnight was designed by Yonatan Sompolinsky and Michael Sutton, who published it in 2022 in the paper The DAG KNIGHT Protocol: A Parameterless Generalization of Nakamoto Consensus. Sompolinsky also co-authored GHOSTDAG and the original GHOST protocol.
What does parameterless mean?
GHOSTDAG needs a fixed parameter, k, chosen in advance from an assumed worst-case network delay. DAGKnight has no such fixed setting. It works out from the DAG itself how much parallelism the honest network is actually producing, and adapts.
Will DAGKnight make transactions faster?
Its main gain is faster, adaptive confirmation. Because it doesn't have to assume the worst-case network delay all the time, confirmations can settle as fast as the real network allows, and slow down safely when the network is struggling. Raw block rate is a separate setting: Kaspa runs 10 blocks per second today, and higher rates are on the roadmap.
Does DAGKnight change mining or the KAS supply?
No. Kaspa stays proof of work with kHeavyHash, and the emission schedule is unaffected. DAGKnight changes how the network orders blocks and decides when transactions are final.
Is DAGKnight as secure as Bitcoin?
The paper proves security against an attacker with less than 50% of the hashrate, the same threshold as Bitcoin's Nakamoto consensus, without relying on a hard-coded latency bound. Like any protocol, real-world security also depends on the implementation, which is still being built and reviewed.
Sources and further reading
- The DAG KNIGHT Protocol, Sompolinsky and Sutton (2022)
- PHANTOM and GHOSTDAG, Sompolinsky, Wyborski and Zohar
- DAGKnight tie-breaking enters code review, Kaspa News (August 2026)
- rusty-kaspa, the Kaspa node source code