Arbitrage Bots Crypto: How to Tell Bot Flow From Real Selling
Open a token chart and the tape fills with red sells. Before you conclude that insiders are leaving, check who is actually on the other side of those prints. A large share of them usually belongs to arbitrage bots crypto traders never notice: automated addresses that buy in one pool and sell in another to capture a price gap. They produce sell prints. They do not reduce anyone's holdings.
That distinction matters, because sell pressure is the number most traders react to, and gross sell volume is the worst possible way to measure it. What you want is net flow, broken down by who is doing the selling.
Gross flow vs net flow: why a red candle lies
Start with the definition. A sell is tokens moving into a DEX pool. A buy is tokens moving out of a pool. That is it. Every swap has two sides, and a chart that colours a print red is only telling you which direction the token moved relative to one pool.
Now consider a transaction that routes tokens through two pools. The token goes into pool A, comes out of pool A, goes into pool B, and settles. Gross volume counts both legs and reports a large sell. Net flow measures the movement per transaction, so a route that passes through two pools nets to roughly zero. Same transaction, two completely different readings.
This is the single biggest source of false sell pressure on a chart. A busy arbitrage bot can print hundreds of red sells in a day while its net position in the token stays close to nothing.
The arbitrage bot pattern
Bots are not hiding. They are just boring to look at, which is why most people skip them. Three tells identify them.
- The address has sent tens of thousands of transactions. Real holders do not do that.
- It holds almost none of the token. It is a courier, not an owner.
- It buys in one pool and sells in another, often inside the same transaction.
That last point is the giveaway. The bot is not expressing a view on the token. It is moving price between two venues that have drifted apart, and it takes the difference. When it sells in pool B, it is because it just bought in pool A. Nothing left anyone's wallet.
If you want a longer treatment of this specific behaviour, the mechanics are worth understanding before you react to the next red candle.
How to check arbitrage bots crypto flow by hand
You can verify all of this yourself without trusting anyone's dashboard. It takes a block explorer and some patience.
Step one: find a repeat seller. Open the token on a block explorer and look at its transfer history. Pick an address that shows up again and again in the same few minutes. Open that address.
Step two: read its profile. Check two numbers: its total transaction count and its current balance of the token. A very high transaction count combined with a near-zero token balance is the bot signature. An address that has been selling for weeks but still holds a large position is a different animal entirely.
Step three: trace one transaction. Open a single transaction from that address and look at the counterparties. If the token went into one pool and out of another in the same call, you are looking at an arbitrage route. If the tokens arrived from a mint or a staking contract before being sold, you are looking at something that does affect supply.
Step four: do it at scale with RPC. Pull Transfer event logs for the token over a block range with eth_getLogs, filter for the pool addresses, and group the senders. On Robinhood Chain this is heavier than you expect: blocks land roughly every 0.1 seconds, so a 24 hour window is about 860,000 blocks. The public endpoint at https://rpc.mainnet.chain.robinhood.com is rate-limited and Robinhood's own docs say it is not for production use. A second public endpoint exists at robinhood-rpc.publicnode.com, and Robinhood's docs recommend Alchemy, Chainstack, QuickNode, Blockdaemon, dRPC, Validation Cloud and GlobalStake as providers. Infura and Ankr were not found to support the chain as of September 2026. If you are doing this often, read the Robinhood Chain RPC guide first.
Doing this by hand for one token is an afternoon. Doing it for every token you are considering is not realistic, which is the actual problem.
How SellTape does it in one message
SellTape is read-only on-chain analysis delivered in Telegram. There is no wallet connection, no signing, no deposits, and it never asks for keys. It currently covers Robinhood Chain (chain id 4663) and Ethereum, with Base, Arbitrum, BSC and Berachain coming.
A report has three parts.
- Contract scan. Upgradeable proxies (EIP-1967, beacon, EIP-1167 minimal proxy); owner and other privileged roles such as owner, treasury, guardian and minter, and whether each is a plain wallet, a Safe with its threshold, a timelock, or a contract; dangerous functions present in the bytecode such as mint, blacklist, pause, fee setters, a trading switch, upgrade and rescue; a simulation of whether the owner can change fees; and a sell simulation from a real holder, where a revert is a honeypot signal. A passing simulation is not a guarantee, since code can be upgraded or fees changed later.
- Holder map. Top holders labelled as pool, staking vault, burn, Safe, contract or wallet, with the share held by staking and pools versus the real float, plus concentration of the top ten real holders.
- Who is selling. Over the last 24 hours across all pools of the token, including Uniswap-style V2, V3 and V4, selling is split into protocol mint-and-sell, treasury-directed selling, unstakers, ordinary holders, wallets not checked for bot behaviour, and arbitrage bots, which are not dumping.
That third part is the one that answers the question this article is about. Instead of one red number, you get the sources.
Every report carries a risk grade of LOW, ELEVATED, HIGH, CRITICAL or INCOMPLETE. Missing data produces INCOMPLETE, never a low grade. Each report also shows the block number, UTC time, caveats, and the line that it is a heuristic and not financial advice. You can see the format on the sample tape and the tiers on pricing.
So who is really selling?
Once bots are separated out, the remaining sell flow is the part that actually changes who owns what. There are four sources worth knowing.
Protocol mint-and-sell
A project contract receives newly minted tokens, a transfer from the zero address, and sells them into a pool. Proceeds often go to the project treasury. This is dilution even when it is by design, and it is invisible on a chart that just shows sells. Note also that the max supply shown on data sites is not the same as minted supply, which is what totalSupply() returns on the contract.
Treasury-directed selling
Tokens move from a treasury address into a pool. Whether that treasury is a plain wallet or a Safe matters. A Safe is a smart-contract wallet with a list of owners and a threshold, and both getOwners() and getThreshold() are public read calls. A Safe with a threshold of 1 and one owner is controlled by a single key, even though it is technically a multisig contract. That is worth checking before you treat treasury selling as a committee decision. More on that in the 1-of-1 Safe explainer.
Unstakers
Tokens leave a staking or vault contract that holds a large share of supply and are sold in the same flow. This is scheduled behaviour in many cases, which means it can be anticipated. If you want to know how these flows line up on a calendar, see token unlocks and unstaking.
Ordinary holders
Real wallets selling real positions. This is the category people assume is everything, and it is usually the smallest of the four once you strip out bots and protocol flow.
Real float changes the picture again
There is a second layer here. Supply held in staking vaults, liquidity pools, burn addresses and treasury Safes is not freely trading. A quoted market cap can badly overstate what can actually be sold. If you are sizing crypto sell pressure, you need the float, not the headline number. The real float and liquidity pools post covers how to work that out.
The tools that do not split sell flow
Plenty of good tools exist in this space. None of them answer the question this article is about, and that is not a criticism, it is a scope note. As of September 2026:
| Tool | What it shows | Splits sell flow by source? |
|---|---|---|
| Token Sniffer | Automated token audit scores for EVM tokens | No |
| Honeypot.is | Honeypot and tax simulation; Ethereum, BSC and Base only | No |
| Bubblemaps | Holder-cluster maps; lists Robinhood Chain as supported | No |
| DexScreener | Charts, pairs and trading data across many chains | No |
| GoPlus | Token-security API flags such as mintable, owner, honeypot-style | No |
| SellTape | Contract scan, holder map, sell-by-source breakdown | Yes |
Honeypot.is does not cover Robinhood Chain, which is worth remembering if that is where you are trading. Bubblemaps shows wallet clusters, not a sell-by-source breakdown. DexScreener shows trades, not who is behind them by category. A comparison of the first three is in the tool comparison post, and if you want to understand what a failing sell simulation actually means, read the honeypot check guide.
What to do with this
The practical takeaway is a change in the order you ask questions.
Do not start with "how much sold today." Start with "who sold today." If the answer is mostly arbitrage bots, the red on your chart is noise and the price is being kept honest across pools. If the answer is a minting contract, a treasury Safe, or a staking vault emptying, that is real supply hitting the market and it deserves your attention.
Then check whether the rest of the setup supports the trade: whether the contract can be upgraded, whether the owner can change fees, whether a sell simulation passes, and how much of the supply is actually float. Those are the inputs that decide whether a sell print is a warning or a rounding error.
Key takeaways
- A sell is tokens into a pool, a buy is tokens out. A route through two pools nets to roughly zero, so gross sell volume overstates pressure.
- Arbitrage bots show a consistent signature: tens of thousands of transactions, almost no token balance, and buys and sells across different pools, often in one transaction.
- They move price between pools. They do not reduce anyone's holdings.
- Real sell flow comes from protocol mint-and-sell, treasury-directed selling, unstakers, and ordinary holders. Each has a different meaning.
- Check the float, not the market cap, and check whether a treasury is a real multisig or a 1-of-1 Safe.
- Missing data should produce an INCOMPLETE grade, never a low one. Treat any tool that grades a token low on absent data with suspicion.
Scan a token free in Telegram at t.me/SellTapeBot.
SellTape is an information service, not financial advice.