Which Chains Stopped? 43 Days of Block Timestamps Across 110 Chains

When a chain halts, nothing errors out. RPC endpoints stay up, eth_blockNumber keeps answering, and the answer stops changing. An app that only checks "is my RPC responding" sees a healthy chain. Its users see transactions that never confirm.

We wanted to know how often this actually happens, so we read block timestamps directly from 110 chains for the 43 days from August 25 to October 6, 2026. Ten chains stopped producing blocks for 10 minutes or more. Four of those stops lasted longer than half a day, and each of the four was a deliberate halt by validators after an exploit.

The full list is published as data: chain-halts-2026.json.

The halts

All times UTC. "Last block" is the last block produced before the stop.

TAC's halt began three days before our window opened, so its length is measured from the actual last block, which we read directly.

"No public report found" means exactly that. We looked for status-page entries and announcements and found none. Short pauses can be maintenance that simply wasn't announced.

Three patterns

The long halts were security decisions, not failures. TAC (exploit report), Neutron, Cosmos Hub and Cronos (Tectonic exploit) all stopped because validators chose to stop them, to keep an attacker from moving funds. The Neutron attack also took down the Cosmos Hub for almost 25 hours, because stolen ATOM had reached the Hub. One incident halted two chains. The Cosmos Hub forum timeline puts the Neutron halt at about 07:44 on September 22; on-chain, Neutron's last block before the halt, 61,635,574, is timestamped 07:43:57.

Restarts stutter. After the long stops, Neutron paused twice more within the hour it came back (19 and 22 minutes), TAC paused for 41 minutes right after its restart, and Cronos stopped three times in 21 hours. If your app reconnects the moment blocks resume, plan for a few false starts.

Planned upgrades can run long. Starknet announced roughly 10 minutes of downtime for its October 6 upgrade to v0.14.4, and its status page marked the upgrade complete at 08:13. On-chain, no block was produced between 08:17:13 and 08:56:09, a 39-minute gap.

What the rest looked like

The other 94 chains we scanned, including Ethereum, Arbitrum, Base, OP Mainnet, Polygon, BNB Smart Chain, Avalanche and zkSync Era, had no gap of 10 minutes or more in the window (six more were excluded, see below). That is a strong record for a period that included a widely exploited Cosmos EVM bug and several high-profile attacks.

How we measured

For each chain we found the first block after August 25 00:00 UTC and the last before October 7 00:00 UTC, read timestamps at 258 evenly spaced heights (one every four hours), and bisected any segment that ran long until we reached the exact pair of consecutive blocks with the gap. A gap counts when it is longer than 10 minutes or 30 normal block times, whichever is larger.

The scan has one blind spot: some sequencers produce a burst of blocks one second apart right after a pause, and that catch-up hides the pause inside a four-hour segment. We found Scroll's September 11 stop a second way, through our own minute-by-minute height samples, then confirmed it on-chain: block 34,992,723 came 24 minutes after the one before it and carried 48 transactions against a usual one or two.

Four judgment calls:

• Some chains only make blocks when there are transactions. On those chains a long gap may just mean nobody transacted. We counted a gap only when the first block after it showed a backlog. Scroll, Taiko and ADI did (15 to 48 transactions after the gap, against 1 to 11 before). Lens didn't, so its 17-minute gap is not listed. Arbitrum Nova, Beam, Kite, Playnance and HAQQ produce blocks this way at low activity, and we left their gaps out entirely.
• Seven chains are not covered because the nodes behind our endpoints no longer serve blocks from late August: Akash, Celestia, Dymension, Filecoin, Injective, Sei and Stable.
• Coverage starts later for Neutron (September 21), Axelar (September 27) and Osmosis (October 5), for the same reason. Neutron's halt is still fully inside its window.
• Non-EVM chains outside the Cosmos SDK (Solana, Sui, Aptos, Polkadot, Bitcoin and others) were not part of this scan.

Detecting a halt in your own code

Liveness checks don't catch halts. Check the age of the newest block instead:

import { createPublicClient, http } from "viem";

const client = createPublicClient({
  transport: http("https://rpc.swiftnodes.io/rpc/cronos?key=YOUR_API_KEY"),
});

const block = await client.getBlock({ blockTag: "latest" });
const ageSeconds = Math.floor(Date.now() / 1000) - Number(block.timestamp);

// Pick a limit well above the chain's normal block time.
if (ageSeconds > 120) {
  console.warn(`No new block for ${ageSeconds}s: chain halted or node stale`);
}

To tell a halted chain from a stale node, ask a second, independent endpoint. If both return the same old block, the chain has stopped. If one is ahead, the other node is behind. Our open-source rpc-doctor does this comparison with --reference.

For finality during and after a halt, see how old "finalized" is on 62 chains and soft vs hard finality on L2s.

The data

chain-halts-2026.json has every halt above with block numbers and UTC timestamps, plus the lists of chains scanned, excluded and not covered. It is free to use under CC BY 4.0; please link to this post. We plan to repeat the scan each quarter.

SwiftNodes serves all 110 of these chains, and more, on one key. Live status is on our status page, and SwiftNodes has a free tier to try it.