From the beginning, Solana broke with precedent.

While most L1s inherited the mempool as a default coordination layer, Solana eliminated it entirely with Gulf Stream. Gulf Stream sends your packet straight to the next leader. No gossip or global queue, in theory no chance for spectators to peek.

In reality, every transaction still passes through a validator on its way to the leader. If a validator wants to, it can watch that flow. They’re not getting the full firehose - but they’re not blind either. That’s how some validators offer “priority services” or monetize partial visibility behind the scenes.

For the average user, it feels mempool-less. But for operators with access and incentives, the shadows are real.

📚If you’re new Solana - jump into our brief history of Solana and a deep dive into the technical milestones that moved the ecosystem forward.

‍

While Solana has no public mempool, it has something much harder to police: information surfaces that leak intent before execution.

This primer is for builders and LPs. Especially those thinking about or using: 

• DEXs
• Perp engines
• Vault strategies
• Liquidity routing

If you’re deploying size or managing latency-sensitive strategies on Solana, this is your threat model broken down for you.

‍

Mempools weren’t designed for MEV. They were designed for buffering.

📚Want to know more about MEV? Learn more about the mechanics of extraction and information asymmetry chipping away at your profits. 

Bitcoin’s way to hold overflow, Ethereum’s place to sort by gas price. Useful? Sure. But also legible. Dangerously so - ​a public waiting room where every trade is laid out like cards on a poker table.

See a pending swap → copy it → jump the queue.

That’s MEV in one breath: information asymmetry before finality.

Solana killed the room. The leaks stuck around.

‍

Gulf Stream fires transactions at the known leader. Nice theory. This design assumes that:

Fast path = less time to spy.

Blockhash expiry = no long-tail simulations

No mempool = no frontrunning

For a while, that worked. Especially in the pre-QUIC, pre-Jito days.

Then the money showed up.
Every coordination mechanism comes under attack the moment there’s money on the line. And Solana’s transaction pipeline, despite its elegance, became too predictable.

QUIC added handshake friction. SWQoS introduced stake weighted lanes. Jito introduced delay-based auction logic. The result:

The mempool didn’t reappear publicly. A shadow one was born.

Solana’s no mempool model is fast - no doubt. But it ain’t flawless. Even without a traditional mempool, congestion still happens.When a validator becomes the leader, inbound traffic can spike above 1 GB per second - as the entire network targets that single node with transaction packets (2024).

Transaction failures are common:

• If a blockhash is too new or from a forked chain, your hash is dropped
• If an RPC pool is out of sync? Dropped
• If the rebroadcast queue exceeds 10,000 transactions? You guessed it - dropped again
• Even if processed, if your transaction lands on a minority fork, it could get invalidated before finalization.

Apps like Mango fight back with custom loops: sendTransaction(maxRetries = 0) and hammer until the lastValidBlockHeight says “stop.” Remember: a 200 OK from sendTransaction only means one RPC heard you.

‍

The relayer isn’t a bug, it’s a feature: bundle auctions, searcher tips, 200 ms to think.

This sliver of time is enough to:
1. Simulate your swap.
2. Quote the edge.
3. Reorder the block.

Jito turned every transaction into a private signal. With 80 %+ stake running Jito (Q1 2024) nearly every Solana trade walked this gauntlet.

April 2024: Jito shutters its public mempool after sandwich attacks got too loud. The 200 ms delay lives on - silent and private.

Translation for traders: your flow is visible, your slippage is target practice, your edge is someone else’s P&L.

‍

Even without a mempool, your intent leaks. Why?

Because metadata shouts:

• Slippage tolerance
• Amount in/out
• Token directionality

Every time a trader submits a swap, they define a slippage window. That tolerance becomes a target.

Telegram-based retail bots often set 10-20% slippage to guarantee execution. They might as well hang neon signs: COME SANDWICH ME.

“You wouldn’t go to a flea market with a sign around your neck saying what you’re willing to pay, before you even start bargaining.”

Even worse: applications that batch transactions or use load-balanced RPCs can expose slippage data across multiple hops - introducing new surfaces for front-running bots to exploit.

‍

We don’t ask Solana to change. We cloak it.

Darklake uses zero-knowledge encryption to protect Solana-native transactions before they hit the network:

• Slippage? Hidden.
• Payload? Ciphertext.
• Simulations? Impossible* 

*Well not entirely impossible - slippage is obscured but validators can still run the transaction to see if it passed or not. What’s important is that trade intent is no longer a thing to observe or extract - it simply doesn’t exist in a parseable form. The transaction becomes a black box. 

Relayer can delay it, but it can’t read it. Searchers can’t price what they can’t parse.

Darklake inserts a layer of unpredictability. It removes not just the surface area, but the very notion of predictable payloads. Because you can’t control who’s watching. But you can choose what they see.

Fast is good. Private is better. Fast and private is the future.

‍