“The Swirlds Hashgraph Consensus Algorithm: Fair, Fast, Byzantine Fault Tolerant” was published in 2016 by Leemon Baird, a computer scientist and co-founder of Swirlds Inc. The paper introduces the Hashgraph data structure and a consensus algorithm built on top of it — achieving Asynchronous Byzantine Fault Tolerance (aBFT), which is the strongest possible security guarantee for a distributed system.

The key insight is that consensus on the order of transactions can be computed independently by every node from the same local information (the hashgraph), eliminating the need to send vote messages. Baird calls this virtual voting.

> PDF hosting: The Hedera/Swirlds whitepaper is available at hedera.com/hh-whitepaper-v2.1. Note: the Hashgraph algorithm is patented by Swirlds; Hedera operates as a permissioned network under an exclusive license.

---

Publication and Context

Leemon Baird (PhD in computer science) spent years on distributed consensus before founding Swirlds with Mance Harmon in 2016. He was motivated by a fundamental observation: every existing consensus algorithm either sacrificed safety (Nakamoto), required synchrony (PBFT), or produced high messaging overhead (many BFT variants).

Asynchronous BFT — consensus secure even under arbitrary message delays, reordering, or temporary network partitions — had been proven theoretically possible but impractical due to the “FLP impossibility” result (Fischer, Lynch, Paterson, 1985). The FLP result shows that deterministic consensus is impossible in a fully asynchronous network with even one faulty node. Hashgraph sidesteps this by being probabilistically finalized (not deterministic), and by operating under a slightly stronger assumption (eventual message delivery).

Key facts:

• Paper: 2016 (Swirlds technical report); updated v2.1 in 2020
• Hedera governance launch: 2019; mainnet public access 2019
• HBAR: Native token for transaction fees and staking
• Governing Council: 39 term-limited organizations (Google, IBM, Boeing, LG, etc.)

---

The Hashgraph Data Structure

The hashgraph is a Directed Acyclic Graph (DAG) of “events.” Each event is a record created by a node when it:

1. Receives and processes a “gossip” from another node
2. Or initiates a gossip spontaneously

Event contents:

• A set of new transactions
• A cryptographic hash of the creator’s previous event (self-parent)
• A cryptographic hash of the other-parent event received in the gossip
• A timestamp and signature

Because every event records the hashes of both parents, each node builds a complete picture of who told whom what and when — the “gossip about gossip.”

---

Gossip About Gossip

In traditional gossip protocols, nodes share information they know. In Hashgraph, nodes share their entire known hashgraph — including all gossip events and their parentage. This means every node quickly learns not just what transactions exist, but the complete history of how information spread through the network.

This means every node can independently reconstruct how any other node would vote on any question, based on the hashgraph structure. No vote messages are necessary.

---

Virtual Voting

Hashgraph consensus is determined by three concepts applied to the DAG:

1. Witnesses: The first event created by a node in each “round” is called a witness.

2. Famous Witnesses: A witness is “famous” if most other nodes’ witnesses in the next round can “see” it (i.e., there is a directed path from them to it via the gossip graph). Famous witnesses serve as anchors for ordering.

3. Consensus Order: An event’s consensus order is determined by the round number of the first round in which famous witnesses that can “strongly see” the event exist. The timestamp is the median of timestamps across all the famous witnesses’ first knowledge of the event.

Every node computes the same witnesses, the same famous witnesses, and the same consensus order — without sending any vote messages. This is why it’s called “virtual voting.”

---

Sections of the Whitepaper

Section	Content
1. Introduction	The consensus problem; aBFT requirements; existing approaches
2. The Hashgraph	Event structure; gossip; the DAG construction
3. Gossip About Gossip	Why full graph sharing enables virtual voting
4. Witnesses and Rounds	Round assignment; witness identification
5. Famous Witnesses	Fame determination via virtual voting
6. Consensus	How order and timestamp are computed
7. Proofs	Formal safety (no fork) and liveness proofs
8. Performance	Expected O(log N) gossip rounds to reach consensus

---

Key Properties

Property	Hashgraph
Fault model	aBFT: up to 1/3 Byzantine nodes, arbitrary message delays
Finality	Probabilistic within seconds; effectively certain within minutes
Fairness	Consensus timestamp is the median real timestamp (hard to manipulate)
Throughput	10,000+ TPS in Hedera’s permissioned setup; 1,000–2,000 TPS typical
Energy	No mining; negligible energy consumption

---

The Patent and Permissioned Structure

The Hashgraph algorithm is patented by Swirlds Inc. Hedera Hashgraph operates under an exclusive license from Swirlds. This creates a fundamental limitation: the Hashgraph algorithm cannot be forked or replicated without Swirlds’ permission. Hedera’s governance model — a council of 39 term-limited enterprises — governs the network, but Swirlds retains patent control.

This is why Hedera is classified as a permissioned public ledger, not a fully open decentralized network. The governing council members rotate, but the validator node software and the consensus mechanism are under corporate IP control.

---

Reality Check

The aBFT guarantee is real and among the strongest in deployed consensus systems. Hedera’s transaction throughput and fees are competitive with centralized services. However, the Swirlds patent and permissioned governing council structure attracted sustained criticism from the decentralization community. Critics argue that a patent-encumbered consensus mechanism fundamentally limits the network’s censorship-resistance and decentralization guarantees.

By 2025, Hedera had processed billions of transactions, primarily in tokenization, supply chain, and micropayment use cases — often outpacing more “decentralized” chains in raw throughput.

---

Related Terms

• Hedera
• Stellar Whitepaper
• Algorand Whitepaper
• Avalanche Whitepaper

---

Research

• Baird, L. (2016/2020). The Swirlds Hashgraph Consensus Algorithm. Swirlds Technical Report.

— Primary source. The formal proofs in §7 are the key academic contribution.

• Baird, L., Harmon, M., & Madsen, P. (2019). Hedera: A Public Hashgraph Network & Governing Council. Hedera Whitepaper.

— The Hedera-specific deployment paper describing the governing council model.

• Fischer, M., Lynch, N., & Paterson, M. (1985). Impossibility of Distributed Consensus with One Faulty Process. JACM.

— The FLP impossibility result that Hashgraph’s design works around.