On the morning of 28 October 2026, swapping a stablecoin for ether cost about 7 cents in network fees on Ethereum mainnet. The same blockchain, where users swap tokens directly from their own wallets without handing custody of their funds to a company.">swap cost a fraction of a cent on Base, Arbitrum, OP Mainnet, BNB Chain and Avalanche, and about five hundredths of a cent on Solana. Yet Ethereum still held 46.2 billion dollars of DeFi deposits that day, more than the next six chains together, according to DefiLlama.
So the cheapest network is not where most of the money sits. Picking the best blockchain for DeFi is a trade-off between four things you can check yourself: what an action costs, how much liquidity is already there, who can change the rules or stop the chain, and how long you wait before a payment is final (or before you can leave). No chain wins all four. This lesson measures each network on those tests with figures dated 28 October 2026, so you can choose a chain for a given job instead of following a ranking.
The previous lesson, how to measure DeFi TVL and read DefiLlama, explained why total value locked is priced in dollars and moves with token prices. Keep that in mind here: Ethereum’s staking or liquidity, measured at current market prices. It is the main size metric in DeFi.">TVL halved in a year, but a large part of that is the price of ether, not people leaving.
Four tests for any DeFi network: fees, liquidity, security, finality
Before comparing chains, a short reminder. A layer 1 is a blockchain with its own validators and consensus, such as Ethereum, Solana or BNB Chain. A layer 2 runs transactions elsewhere and posts its data and proofs back to a layer 1, which it relies on for settlement. Our guide layer 1 vs layer 2 covers the basics; below we only look at what the difference means for DeFi users.
Fees: a swap costs cents almost everywhere in October 2026
The table shows the gas price read from each network’s public node at 09:45 UTC on 28 October 2026 (HyperEVM at 09:57 UTC), and what a simple transfer and a swap would cost at that moment. We assumed 150,000 gas for a swap, a common order of magnitude for a single-pool trade; routed trades through several pools use more. Dollar values use DefiLlama token prices from the same morning.
| Network (28 Oct 2026, 09:45 UTC) | Gas price | Transfer (21,000 gas) | Swap (about 150,000 gas) |
|---|---|---|---|
| Ethereum mainnet | 0.196 gwei | about $0.010 | about $0.07 |
| Base | 0.006 gwei | about $0.0003 | about $0.002 |
| Arbitrum One | 0.020 gwei | about $0.001 | about $0.007 |
| OP Mainnet | 0.001 gwei | about $0.00005 | about $0.0004 |
| BNB Chain | 0.05 gwei | about $0.0008 | about $0.006 |
| Avalanche C-Chain | 2.5 gwei | about $0.0005 | about $0.003 |
| HyperEVM (Hyperliquid) | 0.10 gwei (09:57 UTC) | about $0.0002 | about $0.001 |
| Solana | 5,000 lamports per signature | about $0.0005 | base fee plus any priority fee |
These are gas fees at one instant. They rise when blocks fill, and a busy hour can multiply them. Layer 2s also charge a data fee for posting to Ethereum, which our figures leave out; since Ethereum added cheap blob space for rollups it is usually small. On Solana, the base fee of 5,000 lamports per signature is only the floor, and in busy markets users add a priority fee to get in.
In late 2026, the gap between Ethereum mainnet and the rest is a few cents for an ordinary swap. Fees matter for someone who trades dozens of times a day or moves very small amounts. For most users, they no longer decide the question on their own.
Liquidity: where the money sits is not where it trades
Liquidity decides the price you get. A deep smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool lets you swap 50,000 dollars with little slippage; a shallow one moves against you. Lending markets work the same way: the bigger the pool, the easier it is to borrow or withdraw without hitting a cap. The first measure is total value locked (TVL) by chain, from DefiLlama’s chain rankings (its default series, which leaves out exchange balances).
| Chain | TVL 28 Oct 2026 | TVL 28 Oct 2025 | All-time peak (date) |
|---|---|---|---|
| Ethereum | $46.21 bn | $88.19 bn | $107.46 bn (9 Nov 2021) |
| Bitcoin (as DefiLlama counts it) | $6.36 bn | $8.23 bn | $9.13 bn (7 Oct 2025) |
| Base | $6.13 bn | $5.30 bn | $6.47 bn (5 Oct 2026) |
| Solana | $5.76 bn | $11.78 bn | $13.23 bn (14 Sep 2025) |
| BNB Chain | $5.53 bn | $8.66 bn | $21.78 bn (5 May 2021) |
| Tron | $5.33 bn | n/a | n/a |
| Hyperliquid L1 | $2.49 bn | $2.37 bn | $2.77 bn (19 Sep 2025) |
| Arbitrum | $1.52 bn | $3.65 bn | $4.20 bn (7 Oct 2025) |
| Avalanche | $0.56 bn | $1.94 bn | $11.49 bn (2 Dec 2021) |
| OP Mainnet | $0.43 bn | n/a | n/a |
Ethereum and Solana both lost about half their TVL in a year, while Base reached a record in October 2026 and Hyperliquid held steady. Capital did not leave DeFi wholesale: some of it moved, and the rest was repriced by falling token prices. Arbitrum fell from 3.65 to 1.52 billion dollars over the same period, so the old idea that Arbitrum is the default DeFi layer 2 no longer matches the data.
TVL measures stored value. Activity is a different picture. Over the 30 days to 28 October 2026, DefiLlama recorded the following spot DEX volume and fees by chain (fees here means what users paid to applications and to the chain, gas included).
| Chain | DEX volume, 24 h | DEX volume, 30 days | Fees, 30 days |
|---|---|---|---|
| Solana | $1.85 bn | $60.8 bn | $410 m |
| Ethereum | $1.15 bn | $34.8 bn | $332 m |
| Base | $0.89 bn | $29.8 bn | $53 m |
| BNB Chain | $0.92 bn | $27.5 bn | $78 m |
| Hyperliquid L1 (spot only) | $0.19 bn | $8.0 bn | $157 m |
| Arbitrum | $0.15 bn | $4.7 bn | $20 m |
| Avalanche | $0.06 bn | $3.6 bn | $10 m |
| OP Mainnet | $0.02 bn | $0.8 bn | $5 m |
Solana traded almost twice Ethereum’s spot volume with one eighth of its TVL. Ethereum is where value is parked, in staking and lending; Solana is where it turns over, much of it in fast token trading. Hyperliquid’s main business, perpetual futures, is missing from the spot column because DefiLlama’s free data does not include it, which is why its fees (157 million dollars in 30 days, third among the chains in this table, behind Solana and Ethereum) say more about its size than its DEX volume does.
Outside the networks covered below, DefiLlama’s top 15 on 28 October 2026 also included Tron (5.33 billion dollars, mostly a stablecoin payment rail), Robinhood Chain (1.14 billion), Monad (0.97 billion), Polygon (0.70 billion), Arc (0.63 billion) and Plasma (0.55 billion). Several of these are young; check how long a chain has run before trusting it with size.
Security: who can change the rules or halt the chain
Every chain asks you to trust someone. On a layer 1, that is the validator set: how many independent operators there are, and how many it takes to stop or censor the chain. A useful single number is how many of the largest validators together control more than one third of stake, since in most proof-of-stake designs that share is enough to halt finality. On a layer 2, the question is different: who runs the sequencer that orders transactions, who can upgrade the contracts holding your funds on Ethereum, and whether you can leave without anyone’s permission.
The numbers vary widely. On 28 October 2026 Ethereum’s beacon chain had about 816,000 active validators, a count we took from a public beacon node. Solana had 662 active vote accounts, and the 18 largest of them held more than a third of stake. BNB Chain’s documentation describes an active set of 45 validators, 21 of them in the “Cabinet” that produces most blocks. Hyperliquid’s own API listed 27 active validators, and the three largest, all run by the Hyper Foundation, held more than a third of stake between them.
Finality: from 150 milliseconds to a week
Finality is the moment a transaction can no longer be reversed. For a trader it decides how soon the next move is safe; for anyone sending funds to an exchange or a bridge, it decides how long the other side waits. Ethereum produces a block every 12 seconds, but a block becomes final only after two epochs of 32 slots; ethereum.org puts it at about 15 minutes. Solana now targets about 150 milliseconds (see below). BNB Chain says blocks reach finality in about 0.65 seconds under normal conditions. Avalanche’s documentation says its C-Chain settles in under a second. Bitcoin has no final moment at all: each new block, about every 10 minutes on average, makes a reversal harder, and bitcoin.org calls six confirmations the “minimum recommendation for high-value bitcoin transfers”, about an hour.
Layer 2s add a second clock. Their sequencer confirms a transaction in a second or two, but that is a soft confirmation. Withdrawing to Ethereum through the official bridge on an optimistic rollup takes days, because the system leaves time for anyone to challenge a wrong state. That delay is covered in the layer 2 section.
Ethereum mainnet keeps the deposits, mostly in staking and lending
Ethereum mainnet is where the large, slow money lives. Its top protocols by TVL on 28 October 2026, per DefiLlama, were Lido (23.1 billion dollars), Aave V3 (13.7 billion), SSV (12.4 billion), Binance staked ETH (8.6 billion), EigenCloud (6.1 billion), Sky (5.3 billion), Morpho (5.1 billion), SparkLend (5.0 billion) and Ethena USDe (4.6 billion). That is staking, restaking, lending and stablecoins: positions people hold for months. If you are new to how the network itself works, our lesson what Ethereum is and how it works covers the basics.
Why does the money stay where fees are highest? Depth comes first: the biggest lending pools and stablecoin issuers are here, and a large borrower finds the same size nowhere else. The validator set is also the largest of any chain in this lesson, about 816,000 active validators on 28 October 2026. Layer 2s settle to Ethereum too, so funds there are one step closer to the base layer than funds anywhere else.
The costs are slower finality (about 15 minutes for full finality) and fees that, while low this autumn, still rise faster than elsewhere when demand spikes. On a 20-dollar trade, a 7-cent fee is a third of a percent. On a 200,000-dollar deposit into a lending market, the same cents disappear next to the question of depth. Our guide to the layers of the DeFi stack shows how the applications built on top of this base layer fit together.
Base, Arbitrum and OP Mainnet: the three large layer 2s, all at Stage 1
Among Ethereum’s layer 2s, the order of size changed in 2025-2026. Base, run by Coinbase on the OP Stack software, is now the largest by every measure: 6.13 billion dollars of DeFi TVL on DefiLlama and 15.1 billion dollars of total value secured on L2BEAT, which counts every token held on the chain, used in DeFi or not. Arbitrum One follows with 1.52 billion of TVL (9.82 billion secured), then OP Mainnet, the chain of the Optimism project, with 0.43 billion (1.71 billion secured), all as of 28 October 2026. On Base, most DeFi money sits in lending: Morpho Blue alone held 4.5 billion dollars, against 0.32 billion in Uniswap V3.
All three are optimistic rollups: they post transaction data to Ethereum and assume each new state is valid unless someone proves otherwise during a challenge window. Arbitrum runs a permissionless fraud-proof system called BoLD, deployed in February 2025, and is governed by a Security Council of 12 members, of whom nine must sign an emergency action. Base added a second proof route in 2026, in which a state claim can be backed by a hardware attestation or a zero-knowledge proof.
What L2BEAT’s stages mean for your funds
L2BEAT, an independent research site, grades rollups on a three-step scale. The stages are not a security score; they answer one question: if the people running the rollup turned against you, could you still get your money out?
- Stage 0 means the rollup posts its data to Ethereum and has a working proof system, but its operators can still in practice change the rules. Treat it as a managed service with an Ethereum backup.
- Stage 1 means that, bugs aside, blocking or forging withdrawals requires compromising at least 75% of a Security Council, and upgrades from outside that council leave users at least seven days to exit.
- Stage 2 means anyone can submit fraud proofs, users get at least 30 days to leave before an unwanted upgrade, and the Security Council may only step in for bugs that can be proven on chain.
On 28 October 2026 L2BEAT tracked 94 projects: 65 at Stage 0, 6 at Stage 1, 4 at Stage 2 and 19 for which the scale does not apply. Base, Arbitrum One and OP Mainnet are all Stage 1, along with Starknet, Ink and Unichain. The four Stage 2 projects are tiny, so no layer 2 of meaningful size has reached Stage 2, and some large ones, such as Robinhood Chain (2.35 billion dollars secured) and Mantle (1.31 billion), are still at Stage 0.
Stage 1 still leaves real power with a few parties. On Base, the sequencer that orders transactions is run by Coinbase, and L2BEAT’s page shows upgrades approved jointly by an 8-of-11 Security Council and a 3-of-6 coordinator multisig, with no delay before they take effect. Base’s block production stopped for about 33 minutes on 5 August 2025, a reminder that a single sequencer is a single point of failure for anyone needing to act fast. Your funds were not lost in that halt, but nobody could trade or close a position for half an hour.
The week-long exit, and why Base now says five days
The withdrawal delay is the price of the optimistic design. To move ETH from OP Mainnet back to Ethereum through the official bridge, you start the withdrawal on the layer 2, prove it on Ethereum, then wait out a 7-day challenge period on mainnet before finalising it in a third transaction. Arbitrum’s period is 6 days and 8 hours according to L2BEAT, longer if a challenge is opened. Base cut its standard path from seven to five days with an upgrade of 25 June 2026, according to L2BEAT, with a one-day path when both of its proof systems agree.
So during that week, your money is not lost, just not usable on Ethereum. Third-party bridges and fast-exit services let you skip the wait by paying a fee and taking on someone else’s risk; how they work, and how they fail, is the subject of a later lesson. For deposits in the other direction, from Ethereum to a layer 2, there is no week-long wait: funds usually arrive within minutes.
Solana: the busiest trading venue, now with 150-millisecond finality
Solana is a single fast chain with no layer 2s in the Ethereum sense. On 21 October 2026, at about 13:36 UTC, it switched its consensus to a new protocol, Alpenglow. The Solana Foundation’s upgrade page says Alpenglow is live on mainnet since epoch 1064, with finality targeted at roughly 150 milliseconds, down from about 12.8 seconds under the previous TowerBFT design. The same page notes that slot times were cut separately, from 400 to 200 milliseconds, in stages completed in October 2026. Only the first phase, Votor, has shipped; the second, Rotor, which changes how blocks spread across the network, is not yet scheduled.
The 150 milliseconds is a target stated by the foundation one week after activation, not a measured long-run figure. The switch also needed most operators on board: the foundation says the first Alpenglow block was certified with votes from 446 of the network’s 654 validators.
In DeFi terms, Solana is the trading floor. It led all chains on spot DEX volume in the 30 days to 28 October 2026 (60.8 billion dollars) and on fees (410 million). Its TVL of 5.76 billion dollars is spread across staking (Sanctum validator liquid staking tokens 1.8 billion, Jito 1.1 billion), lending (Kamino Lend 1.3 billion, Jupiter Lend 1.0 billion) and trading (Raydium AMM 1.1 billion), per DefiLlama.
The trade-off is concentration. With 18 validators holding more than one third of stake on 28 October 2026, it would take far fewer parties to halt Solana than Ethereum.
BNB Chain and Avalanche: one large and exchange-linked, one shrinking
BNB Chain is closely tied to the Binance exchange and runs a Proof of Staked Authority design: an active set of 45 validators, 21 of them in a “Cabinet” that produces most blocks while the other 24 produce far fewer, according to BNB Chain’s documentation. The Fermi hard fork of 14 January 2026 cut the block interval from 0.75 to 0.45 seconds, and a BNB Chain post of 20 August 2026 says blocks become final in about 0.65 seconds under normal conditions.
With 5.53 billion dollars of TVL and 27.5 billion of DEX volume in 30 days, BNB Chain is the third or fourth DeFi venue by most measures. Its activity centres on PancakeSwap (1.55 billion dollars in its AMM plus 0.66 billion in V3), the Venus lending market (1.21 billion) and Lista’s lending and liquid staking (about 1.7 billion together), according to DefiLlama on 28 October 2026. The small, permissioned validator set is the main thing to weigh against low fees and fast finality.
Avalanche has shrunk the most. Its C-Chain, the EVM chain most users mean, held 0.56 billion dollars on 28 October 2026, against 1.94 billion a year earlier and 11.49 billion at its peak in December 2021. Aave V3 (0.30 billion), Grove (0.26 billion) and Benqi’s staking and lending are the main protocols left. Avalanche’s documentation says the C-Chain settles in under a second. The network’s focus has moved to separate Avalanche L1s, each with its own validators, gas token and fee market; those are mostly not general DeFi venues.
Hyperliquid is its own layer 1, not a rollup
Hyperliquid started as a perpetual futures exchange and became a chain. It runs its own consensus, HyperBFT, and settles nothing to Ethereum, so it is neither a layer 2 nor a rollup; L2BEAT lists it under “Other” with stages not applicable. Inside one chain state there are two parts: HyperCore, which holds margin accounts and the on-chain order books, and HyperEVM, a general EVM where smart contracts run. The Hyperliquid documentation says HYPE is the native gas token on HyperEVM.
Its DeFi TVL was 2.49 billion dollars on 28 October 2026, per DefiLlama. That figure leaves out the USDC traders deposit as margin through the Hyperliquid Bridge (6.31 billion dollars, which DefiLlama files as a bridge). The largest lines inside it were Hyperliquid Earn (1.48 billion dollars of lending), Kinetiq’s kHYPE liquid staking (0.85 billion) and the HyperLend pool (0.32 billion). Its 30-day fees of 157 million dollars came mostly from futures trading.
The speed is that of an exchange: Hyperliquid’s documentation gives a median end-to-end order latency of 0.2 seconds for a client located close to its servers. The cost is a small validator set. Its API showed 27 active validators on 28 October 2026; the five run by the Hyper Foundation held 47% of active stake, and the three largest alone held more than one third. Anyone using it should treat it as a fast venue with concentrated control, not as a settlement layer with Ethereum-like decentralisation.
DeFi on Bitcoin: most of it happens somewhere else
DefiLlama credits Bitcoin with 6.36 billion dollars of TVL on 28 October 2026. That figure needs reading carefully. Bitcoin’s base layer has no general smart contracts, so you cannot run a lending pool or an automated market maker on it the way you do on Ethereum. Most of what DefiLlama counts on the Bitcoin chain is BTC held by custodians or protocols so that a token can be issued on another chain, plus BTC locked for staking. The biggest Bitcoin DeFi does not take place on Bitcoin.
Lightning moves payments, not loans
The Lightning Network is a network of payment channels anchored on Bitcoin, built for near-instant, cheap transfers between participants. Its contracts lock funds for a payment and release them against a secret; they do not run lending or trading logic. Public capacity stood at about 3,795 BTC across 32,678 channels and 16,242 nodes in mempool.space’s latest statistics, dated 30 August 2026. It is useful, but it is payments, not DeFi.
Wrapped bitcoin: WBTC, cbBTC and tBTC
The main way bitcoin enters DeFi is as a wrapped token: someone holds BTC and issues a token worth one BTC on Ethereum, Base or Solana, where the lending markets are. WBTC’s bitcoin is held by the custodian BitGo and its partners. cbBTC is issued by Coinbase. tBTC, from the Threshold Network, replaces the single custodian with a randomly selected group of operators and assumes an honest majority among them. In each case, your coin is only as good as the custodian or the group holding the real BTC.
Babylon stakes bitcoin without moving it
Babylon takes another route. Holders lock BTC in a script on Bitcoin itself that they can spend back after a timelock, and delegate it to “finality providers” that help secure proof-of-stake chains. Babylon’s documentation stresses that the BTC is neither wrapped nor bridged; misbehaviour can be punished through one-time signatures and a covenant committee that can slash part or all of the stake. DefiLlama counted 3.30 billion dollars in Babylon on 28 October 2026. This is staking to secure other chains, not lending or trading.
Stacks and sBTC: a separate layer with a signer peg
Stacks is a separate chain that anchors to Bitcoin and runs smart contracts. Its sBTC token is pegged one to one with BTC held in a single Bitcoin address controlled by a multisig of signers, initially 15 community-chosen signers according to Stacks’ documentation. Stacks DeFi is small: 0.06 billion dollars of TVL and 107 million dollars of DEX volume in the 30 days to 28 October 2026.
For a reader holding BTC, DeFi with bitcoin is possible, but it almost always means trusting a custodian, a signer group or a staking protocol, and it mostly happens on another chain. The amounts at risk deserve the same checks as any bridge.
Best blockchain for DeFi: the networks side by side
The table pulls the four tests together for each network, with figures as of 28 October 2026. Read it by the job you want done, not from top to bottom.
| Network | Swap fee (09:45 UTC) | DeFi TVL | Who can stop or change it | Finality / exit |
|---|---|---|---|---|
| Ethereum | about $0.07 | $46.2 bn | about 816,000 active validators | about 15 min |
| Base | about $0.002 | $6.1 bn | Stage 1; Coinbase sequencer; upgrades with no delay by council plus a second multisig | seconds (soft); 5-day exit, 1-day fast path |
| Arbitrum One | about $0.007 | $1.5 bn | Stage 1; 12-member Security Council | seconds (soft); 6 days 8 hours exit |
| OP Mainnet | about $0.0004 | $0.43 bn | Stage 1 | seconds (soft); 7-day exit |
| Solana | about $0.0005 base fee | $5.8 bn | 18 validators hold over 1/3 of stake | about 150 ms target |
| BNB Chain | about $0.006 | $5.5 bn | 45 active validators, 21 in the main block-producing set | about 0.65 s (BNB Chain) |
| Avalanche C-Chain | about $0.003 | $0.56 bn | not measured here | under 1 s (docs) |
| Hyperliquid | about $0.001 (HyperEVM, 09:57 UTC) | $2.5 bn | 27 validators; 3 foundation nodes over 1/3 | 0.2 s median order latency |
| Bitcoin | not comparable | $6.4 bn, mostly custodial or staked | no general DeFi on the base layer | no fixed point; six blocks (about an hour) for large transfers, per bitcoin.org |
The deepest lending markets sit on Ethereum mainnet, where finality takes about 15 minutes. Base and Arbitrum run many of the same applications for a fraction of a cent, but leaving them through the official bridge takes days and a single sequencer orders every transaction until then. Solana and Hyperliquid confirm fastest, and fewer parties could stop them. BNB Chain pairs low fees and sub-second blocks with a 45-member validator set. Every figure in the table is dated 28 October 2026 and will move; check DefiLlama and L2BEAT again before relying on it.
Taxes for US and UK readers when you change networks
Moving to another chain is often more than a click. In the United States, the IRS treats digital assets as property and lists exchanging one digital asset for another among the transactions you must report. A swap on any chain is therefore a disposal with a gain or loss to calculate, and brokers have reported gross proceeds on Form 1099-DA for transactions since 1 January 2025. Our guide to crypto taxes in the United States walks through the calculation.
In the United Kingdom, HMRC’s cryptoassets manual counts exchanging one token for a different token as a disposal for Capital Gains Tax. Wrapping a coin or moving it through a bridge is a grey area in both countries: neither page we checked addresses it directly, so keep records of every step and ask a tax adviser when the amounts matter.
Which network should you start DeFi on?
There is no single answer: the best blockchain for DeFi is the one that passes the four tests for the job you have in mind. Start from that job. Here is a practical order of questions:
- What will you do first? Trying a swap or a small deposit needs low fees; a large, long-term deposit needs depth and security more than cheap gas.
- Can you get there cheaply? Check that your exchange supports withdrawals on that exact network, so you do not pay a bridge on day one.
- Who could freeze or change the system? Look up the L2BEAT stage for a layer 2, or the validator count and concentration for a layer 1.
- How fast must you be able to leave? If you might need the funds on Ethereum at short notice, remember the 5 to 7 days of an optimistic rollup’s official exit.
- Is the liquidity real? Compare TVL and 30-day volume on DefiLlama for the protocol you plan to use, not just the chain.
Practise first on a test network, where tokens have no value and mistakes cost nothing; Ethereum, Base, Arbitrum and Solana all run public testnets with free faucets. When you move to real money, start with an amount you can afford to lose and check the fee before you sign. Our DeFi prerequisites checklist lists the wallet and security steps to take before that first transaction.
Mistakes that cost money when picking a network
Most losses at this stage are not hacks. They come from sending funds to the right address on the wrong network: an EVM address looks the same on Ethereum, Base, Arbitrum and BNB Chain, but an exchange that does not support a network may not credit a deposit sent on it. Check the network name on both sides before sending, and send a small test amount first.
Other common errors: following a link to a fake bridge or a fake front end from a search ad or a direct message; assuming a layer 2 withdrawal arrives in minutes because the deposit did; and comparing chains on claims such as transactions per second, which say nothing about who controls the network. Treat any single headline number with suspicion, including the ones in this lesson: check the date.
The next lesson of the DeFi course looks at what happens between these networks: native bridges, third-party bridges, cross-chain messaging and intent-based systems, and how each of them has failed in the past.
