Origin story
Kaspa's launch structure was not chosen. It was what was left after research into faster proof-of-work ordering ran through DAGLabs, an April 2021 testnet, and a hardware plan, a presale, and a DeFi pitch that each fell apart.
Launch thesis
Every earlier structure would have made Kaspa answer to something outside the protocol: a hardware deal, a presale contract, a startup balance sheet, or a supporting role under someone else's apps. The fair launch proved nothing about later adoption, price, or whether the tech would work. It took the premine, the pre-sale, and the insider table off the starting line.
Research first
Bitcoin at high block rates has one weak point: two honest miners can solve a block seconds apart, and the single-chain rule throws one of those blocks' work away. Kaspa's research line starts from that exact waste. A blockDAG keeps both blocks visible instead of dropping one, then uses a rule like GHOSTDAG to order that graph into one shared payment history.
DBLP lists Yonatan Sompolinsky and Aviv Zohar's 2018 paper, PHANTOM: A Scalable BlockDAG Protocol, as an IACR ePrint paper. Kaspa.org traces Kaspa back through the same blockDAG research line: SPECTRE, then PHANTOM, then the protocols running Kaspa today.
A July 2017 Epicenter episode with Zohar and Sompolinsky describes SPECTRE as dropping the blockchain data structure so blocks can be made all the time and merged after, rather than racing for one slot.
DAGLabs and testnet
Public sources disagree on when DAGLabs was founded. An April 26, 2021 Investing.com launch article says 2017, by Yonatan Sompolinsky and Guy Corem, with $8 million raised from Polychain and Accomplice. The HackerNoon interview intro says Sompolinsky founded DAGLabs in 2018 to build a business on DAG protocols. Either way, DAGLabs paid for and ran part of the pre-mainnet path before Kaspa became a public, community-mined network.
Polychain belongs in the pre-mainnet DAGLabs funding story, and it cuts both ways. Kaspa had no token premine. It also had a VC-funded DAGLabs backstory that shaped the hardware, mining, and fair-launch debate before mainnet existed.
Guy Corem's April 13, 2021 testnet note calls Kaspa a Golang build of Phantom/GHOSTDAG and a generalization of Nakamoto Consensus. The April 2021 Investing.com article lists the testnet bundle: full node, web wallet, faucet, Katnip block explorer, graph inspector, Kashboard. It framed the project as testnet first, mainnet later, with no pre-mining and no founder allocations planned.
Yonatan's Hashdag archive describes the original Kaspad codebase as an adaptation of the Bitcoin client btcd, written in Golang. That holds up; the unsourced "Bitcoin Core 0.19" claim floating around does not.
Could-have-been paths
Hashdag's 2020 community-launch archive said Kaspa would have no premine or founder rewards, In the same breath it called DAGLabs a for-profit miner with a pre-launch ASIC-tuning path and a DeFi-contract auction for rights to future Kaspa coins. Yonatan later said that path never happened: the optical ASIC chips it leaned on weren't ready.
The April 2021 launch article framed Kaspa as faster ordering rails for DeFi apps running on Ethereum. The later Uphold interview recap keeps the same theme underneath: Kaspa's specialty is sequencing, which matters most where transaction order and MEV move real money. That's the source for the "base sequencing layer for another ecosystem" idea. It is not a claim that Kaspa ever shipped as an Ethereum L2.
Yonatan later called the gamenet attempt flawed: closer to a testnet with incentives bolted on than an actual mainnet.
In the HackerNoon interview, Yonatan called DAGLabs a failed attempt to square venture-backed and fair-launch models. Investors pushed for launch with no business model or ROI plan. Asked directly if Polychain pushed him to launch, he agreed that was one way to describe it, and added that Gadi pushed too, because he saw the promise.
Launch
Kaspa.org's current lore page puts mainnet launch in November 2021: a fair-launched Proof-of-Work network with no premine, insider allocation, or pre-sales. The homepage adds the genesis proof: every coin traces back by cryptography to an empty genesis. Mining made the whole supply.
No premine means no official token allocation before mining started. It does not mean equal access to early mining.
Yonatan later estimated Kaspa's core developers mined about 2.5% of total supply after launch. The Kaspa Wiki prehistory page puts the related DAGLabs/Polychain estimate below 3%, roughly 840 million KAS, and says the tighter number is probably closer to 2.5% and under 850 million KAS. That's early-miner context, a separate question from whether there was a premine.
Yonatan's own archive is blunter than the lore. He calls the launch messy, says he baked several bad ideas into the first gamenet rules, and describes a few thousand CPUs finding the project within days, pushing hashrate up no matter what the plan wanted. The Black Tuesday post puts it in one clause: the gamenet idea never caught air, so he swapped the random block reward for a flat one.
Early stress
On November 23, 2021, Yonatan published "Kaspa (Black Tuesday)" after a network crash in the 48 hours before. His account: the developers were not rolling back to an earlier state they liked better, they were rebuilding and comparing a certainly-valid UTXO commitment after a memory problem. That's the sourced version of what gets passed around as the "two-week bug" story.
The repair is in the code, not only in the blog post. Kaspad's November 2021 hard-fork commit is titled "new genesis with the utxo set of the last block," and the mainnet genesis block rusty-kaspa still ships carries a DAA score of 1,312,860 marked as a checkpoint score, plus the same UTXO commitment hash the Black Tuesday post published. Kaspa restarted from the last valid state instead of from an empty ledger. That genesis is timestamped November 22, 2021, the day before the post, which puts the outage in days rather than the two weeks the retelling usually claims.
A dust attack on Yom Kippur, September 2023, was a separate event from the November 2021 launch bug. Kaspa researcher Shai Wyborski's post-mortem gives the mechanism: each attacking transaction spent one UTXO and created 238, so every block added about 1,185 entries to a set where each one costs roughly 85 bytes. That is near 100 KB of permanent state per block, about 5 GB a day. He counted 57 million attacker UTXOs created before September 28 and put the contained damage at "5GB of junk UTXO data we will have to carry permanently." The lasting answer was not a patch but KIP-9's extended mass formula, which prices state growth into transaction mass so the same attack costs the attacker instead of the network.
Rust and Crescendo
Hashdag's 2022 "Kaspa where to" archive says Michael Sutton pushed for a big refactor and Rust rewrite, because the Golang codebase carried technical debt that made it hard for new contributors to work in. Kaspa.org now sums that up as a full rewrite into Rust, paid for by a community grant.
The rusty-kaspa repo describes the Rust build as a drop-in replacement for the older Golang node. Its README puts the Crescendo hard fork on May 5, 2025, moving the network from 1 BPS to 10 BPS. The v1.0.0 release notes name the exact trigger: mainnet DAA score 110,165,000.
Toccata boundary
"Toccata" gets pinned on 10 BPS often enough to be worth fixing. Crescendo is the fork that made 10 BPS live on mainnet. Toccata is a separate, later fork: base-layer programmability with covenants, Silverscript, ZK-facing proof checks, sequencing commitments, and native-asset groundwork, activated at DAA score 474,165,565.
Rusty Kaspa v2.0.1 is the current Toccata release; v2.0.0 shipped the activation. Wallet support, indexer support, and app usage are still building: check Toccata Status for where each stands.
What to say
| Topic | Use | Avoid |
|---|---|---|
| Code origin | Original Kaspad was Golang and source-backed as an adaptation of btcd. | Bitcoin Core 0.19, unless a durable source is added. |
| DAGLabs | DAGLabs paid for and ran pre-mainnet work; public sources differ on 2017 versus 2018 dating. | A single exact founding date with no caveat. |
| Polychain | Polychain belongs in the VC-funded DAGLabs backstory. Premine and pre-sale claims need their own evidence on token allocation. | Ignoring Polychain context or treating it as proof of a premine. |
| Fair launch | No premine, insider allocation, or pre-sales. Every coin was mined, including the DAGLabs/Polychain early mining estimate after launch. | Premine wording for DAGLabs or Polychain, a perfect spread of coins, or no early contributor mining. |
| Business failure | Hardware, presale, and startup paths all fell through before the plain fair launch. | Bankruptcy language unless a stronger source says bankruptcy. |
| Black Tuesday | A November 2021 crash, repaired by a hard fork whose new genesis carried the UTXO set of the last valid block. | A two-week halt. The commit, the genesis timestamp, and the post itself all put the outage in days. |
Sources