I run a buy bot for Solana token teams. Every buy of a token becomes a card in their Telegram group: amount, tokens received, market cap, new holder or not.

The first question people ask when they compare it with Solscan: "why isn't it the same number?"

Sometimes it's exactly the same. Sometimes it's off by a few dollars. Once, my bot was flat-out wrong. Here's every case I've hit, and what I do about each.

1. Paid in a stablecoin: the numbers match exactly

Start with the easy one. This buyer paid in USDT:

A buy card next to the same transaction on Solscan: 2,830 USDT for 100,268 POPCAT

Card: 2,830 USDT. Solscan: 2,830 USDT. When the buyer pays in a stablecoin, there's no price conversion, so there's nothing to disagree about.

(Look at the end of Solscan's line: "on Jupiter Order Engine". This buy was filled by a market maker, the tricky case in section 4. It still shows up as one buy, with the right amounts.)

Everything below is about the cases where there is something to convert, or something extra in the transaction.

2. Paid in SOL: the dollar value moves after the trade

A buyer spends 26 SOL. My bot priced it at $2,469. The next day, Solscan showed the same transaction at $2,488.

Neither is wrong. The bot converts SOL to dollars at the moment of the trade. Solscan's summary shows the value at today's price. SOL moved overnight, so the dollar figure moved too.

Rule: to compare a bot with an explorer, compare the SOL and token amounts, not the dollars. Those must match exactly.

3. The wallet app took a fee: the bot shows more than Solscan

A buy of $MAX: my card said 298.33 USDC, Solscan's summary said 297.00 USDC. A 0.45% difference.

Many trading apps and wallets take their fee inside the same transaction as the swap: a separate transfer from the buyer to the app's fee account, next to the swap itself. The buyer's balance goes down by swap + fee. Solscan's summary describes the swap. My bot measures what left the buyer's wallet.

I kept it that way on purpose: for a buy alert, "what this person actually spent" is the honest number. The swap amount alone hides the cost of buying.

4. RFQ fills: phantom sells

This one was a real bug.

Some swaps aren't filled by a pool but by a market maker (RFQ orders, like Jupiter's JupiterZ). The buyer gets tokens, and the market maker's token balance goes down.

My first logic was simple: a wallet whose token balance drops is a seller. So every RFQ buy also produced a "sell" alert, for a sale nobody made.

The fix: when two wallets in the same transaction show mirror-image changes (one gains the tokens the other loses, for the same payment), the second one is the market maker filling the order, not a seller. The bot drops that side. Since then, an RFQ buy is one buy alert, and nothing else.

5. One buy across several pools: watch the token, not a pool

A token often trades on several pools at once (pump.fun's AMM, Raydium, Meteora). A router can split one buy across two of them.

If your bot watches a single pool, it sees half the buy, or misses it entirely. My bot watches the token itself, and computes the net change per wallet from the transaction's balances. However the buy was routed, it shows up as one alert, with the full amount.

6. Transactions your RPC call refuses to decode

The sneakiest one: buys that never show up at all.

When you fetch a transaction with getTransaction, the RPC only returns versioned transactions if you tell it which versions you can handle, with maxSupportedTransactionVersion. If the transaction is newer than what you asked for, the call returns an error instead of the transaction. If your bot logs that error quietly and moves on, those buys simply vanish.

This bit me in real life. My bot asked for version 0, which was enough for years. Then version 1 transactions started showing up on mainnet, and every one of them came back as:

Transaction version (1) is not supported by the requesting client. Please try the request again with the following configuration parameter: "maxSupportedTransactionVersion": 1 
params = [ signature, { "encoding": "jsonParsed", "commitment": "confirmed", # The highest transaction version your parser handles. # Ask for less and newer transactions come back as an error, not a transaction. "maxSupportedTransactionVersion": 1, }, ] 

Set it to the highest version your parser actually handles, and log loudly when a transaction can't be decoded. A missed buy is worse than a wrong one: nobody even knows to complain.

The checklist

If you build anything that reports trades on Solana:

  1. Compare token amounts, not dollars, when you test against an explorer.
  2. Decide what your number means: what was swapped, or what the buyer spent. Then say so.
  3. Watch out for RFQ fills: a balance going down isn't always a sale.
  4. Watch the token, not a pool.
  5. Never fail silently on a transaction you can't decode.

See it running

The bot posts real buys live, 24/7, on a public channel: t.me/opsnode_live. Pick any alert and check it on Solscan.

Full disclosure: it's my product, Opsnode. If you run a token's Telegram group, there's a free preview of what your own buys would look like.

Have you seen a case I missed? I'd like to know.