One bug from a halt
How much of each blockchain runs on the same software, and why that decides whether a single bug can stop it.
Why one program can stop a chain
In plain words: a blockchain is only as diverse as the software its validators run.
What the thresholds mean
Above one-third
Proof-of-stake chains need about two-thirds of validators to agree. If software run by more than a third of the stake crashes, the rest cannot reach two-thirds: Sui, Cosmos, Aptos and similar chains stop making blocks, and Ethereum keeps making blocks but stops finalizing them.
Above two-thirds
A client this big can do worse than stop the chain. If it has a bug, it can finalize a wrong chain on its own. Validators running it can then be stuck on that chain or slashed when they try to leave.
Single implementation
Every validator runs the same program, so one bad release can halt everyone at once, however many validators there are. Most chains are in this group. Ethereum, Solana and Bitcoin are the main exceptions.
Proof-of-work
Bitcoin, Litecoin and Dogecoin have no voting threshold. Miners keep making blocks, so a software bug splits the chain in two instead of stopping it.
Chains with more than one client
Only a few chains have competing software, and for most of them the mix can be measured from public data. Refreshed twice a day.
| Chain | Largest client | Software mix | What it means |
|---|---|---|---|
| Ethereum Execution layer | Nethermind 42.9% Over one-third | Nethermind 42.9%Geth 42.8%Besu 8.4%Reth 3.3%Erigon 2.7% | Nethermind and Geth are each above one-third: a crash bug in either one alone would stop Ethereum from finalizing until fixed. Blocks would keep coming. Covers about 59% of validators, those that reveal their client. supermajority.info via clientdiversity.org, 15 Nov 2026 |
| Ethereum Consensus layer | Lighthouse 47.1% Over one-third | Lighthouse 47.1%Nimbus 19.1%Prysm 17.6%Teku 6.4%Unknown 3.2%Lodestar 2.7%Erigon 2.3%Grandine 1.6% | Lighthouse is above one-third: a crash bug in it alone would stop Ethereum from finalizing until fixed. Blocks would keep coming. Miga Labs via clientdiversity.org, 15 Nov 2026 |
| Solana By active stake | Agave codebase 81.5% Over two-thirds | Agave and Agave-based (Jito, BAM, Harmonic, Rakurai) 81.5%Firedancer family (incl. Frankendancer, Harmonic, FireBAM) 18.1%Other / unknown 0.4% | Jito, BAM, Harmonic and Rakurai builds are modified Agave, so one Agave bug can hit them all. Frankendancer, counted with Firedancer, still runs parts of Agave, so the real Agave share is higher. Solana halts if more than one-third of stake stops voting. Stakewiz validator list, share of active stake, 15 Nov 2026 |
| Bitcoin Reachable nodes | Bitcoin Core 89.9% Split risk, not halt | Bitcoin Core 89.9%Bitcoin Knots 9.8%Other 0.3% | Proof-of-work has no one-third or two-thirds threshold: miners keep making blocks, so a consensus bug splits the chain instead of halting it, as in March 2013. Knots is a patched Bitcoin Core and shares most consensus code. This counts nodes, not mining power. Bitnodes crawl of 18,862 reachable nodes (user agents), 15 Nov 2026 |
| BNB Chain | BSC client (Geth fork) One dominant client | Share not published | Validators do not publish which client they run. Almost all run the official bsc client, a Geth fork; a Reth-based client exists, but its share cannot be measured from public data. Official repositories and docs, checked 12 Nov 2026 |
| Polkadot | Polkadot SDK node (Parity) One dominant client | Share not published | Most validators run Parity's node built from the Polkadot SDK. Independent clients exist (KAGOME, Gossamer), but validators do not publish which one they run, so their share is unknown. Official repositories and docs, checked 12 Nov 2026 |
| Zcash | zcashd and Zebra Split risk, not halt | Share not published | Zcash has two full-node implementations: zcashd (being retired) and Zebra from the Zcash Foundation. Proof-of-work: a disagreement between them would split the chain rather than halt it. Their share is not published. Official repositories and docs, checked 12 Nov 2026 |
One sequencer: rollups
On these rollups a single company runs the only block producer. Client diversity barely matters here: if that sequencer or its software fails, blocks stop for everyone, as Base did twice in June 2026. Users can still force transactions through Ethereum, but slowly.
- Base OP Stack, run by Coinbase
- Arbitrum One Nitro, run by Offchain Labs
- OP Mainnet OP Stack, run by OP Labs
- ZKsync Era ZK Stack, run by Matter Labs
- Scroll l2geth, run by Scroll
- Starknet Starknet, run by StarkWare
- Linea Linea, run by Consensys
One program for every validator
Most chains have a single production client. One bad release can stop all validators at once, however many there are.
| Chain | Software | Notes |
|---|---|---|
| Polygon PoS | Bor + Heimdall | Validators run Polygon's Bor (a Geth fork) for blocks and Heimdall for checkpoints and milestones. There is no independent second client in production. |
| Avalanche C-Chain | AvalancheGo | All validators run AvalancheGo, maintained by Ava Labs. |
| Sui | sui-node | All validators run the same sui-node software from Mysten Labs, so a bug in one release can stop the whole network. That is what happened in the 2024–2026 halts listed below. |
| Aptos | aptos-node | All validators run aptos-core, maintained by Aptos Labs. |
| TRON | java-tron | The 27 super representatives run java-tron, the only production client. |
| Cosmos Hub | Gaia (CometBFT) | Validators run gaiad on CometBFT. There is one implementation. |
| Osmosis | osmosisd (CometBFT) | Validators run osmosisd on CometBFT. There is one implementation. |
| Celestia | celestia-app (CometBFT) | Validators run celestia-app on CometBFT. There is one implementation. |
| Cardano | cardano-node (Haskell) | Block producers run the Haskell cardano-node. Alternative nodes (Amaru, Dingo) are in development, not in block production. |
| TON | TON node (C++) | Validators run the reference C++ node from the TON core team. |
| XRP Ledger | rippled | Validators run rippled, the only production server. |
| Litecoin | Litecoin Core | Nodes and miners run Litecoin Core. Proof-of-work: a bug would split the chain rather than halt it. |
| Dogecoin | Dogecoin Core | Nodes and miners run Dogecoin Core. Proof-of-work: a bug would split the chain rather than halt it. |
| Hyperliquid | hl-node | Validators run Hyperliquid's own node software, which is not open source. |
| Monad | Monad client | Validators run the Monad client from Category Labs, the only implementation. |
It has already happened
- Ethereum, 4 Dec 2025: the Prysm bug after Fusaka. Voting participation fell to ~74.7-75% (sync participation 75%) at epoch 411448, from >99% normally; missed-slot rate 18.5%; ~41-42 epochs affected; ~382 ETH in attestation rewards lost. Not lost: participation stayed ~9 points above the 2/3 threshold. Prysm's postmortem says a client with more than 1/3 share would have caused a temporary loss of finality. Source
- Sui, 28–30 May 2026: three halts in 48 hours after v1.72. A gas-charging bug in the new address-balances feature crashed validators twice; the restart to fix it then hit a second bug in randomness setup. Every validator runs the same sui-node, so each bug stopped the whole network. Sui postmortem
- Solana, 6 Feb 2024: five-hour halt. A bug in the Agave validator (v1.17) left nearly all validators stuck in a loop. Firedancer was not yet in use on mainnet. Source
How this works and where the data comes from
Ethereum execution and consensus shares come from clientdiversity.org data (supermajority.info and Miga Labs). Solana shares are by active stake from the Stakewiz validator list, grouping builds by the code they are based on. Bitcoin shares count reachable nodes by their user agent in the latest Bitnodes crawl. For chains with no public census we state what the official repositories and docs show: usually a single implementation. Thresholds follow ethereum.org's client-diversity page.
Sources: clientdiversity.org; Stakewiz; Bitnodes; ethereum.org: client diversity
Information only, not financial advice. Data comes from public sources and can be delayed or wrong; check the primary source before acting.