On this page
- From a block to a dashboard
- Why an RPC alone cannot do it
- The common metrics, and how each misleads
- Wallet profit, done properly
- Wallet behaviour analysis
- Snipers and bundles
- What on-chain data cannot show
- Building a small one yourself
- Where AI fits
- BLIX as an example of the pipeline
- The BLIX community
- More on Solana and on-chain data
From a block to a dashboard
| Stage | What it holds | Example |
|---|---|---|
| 1. Raw chain | Blocks of transactions; accounts as bytes | A transaction with twelve instructions and a list of balance changes |
| 2. Decoded | The same, translated into events using each program's layout | "Wallet A bought 3.4M of token X for 0.5 SOL on this pool" |
| 3. Indexed | Events stored in tables keyed by token, wallet and time | All trades of token X, in order; all trades by wallet A |
| 4. Metrics | Numbers computed over those tables | Holders, top-10 share, volume, each wallet's profit |
| 5. Presentation | Screens, alerts, APIs, or an AI reading the numbers | A holder table, a chart, a warning |
Stages 2 and 3 are where the work is. Every program stores its data in its own format, formats change when programs are upgraded, and the chain produces thousands of transactions a second. The system that does this continuously is called an indexer.
Why an RPC alone cannot do it
An RPC node answers "what is in this account now?" and "show me this transaction". It has no way to answer "which wallets bought this token in its first ten seconds?", because that requires having read every transaction of that token and kept the result. You can fetch a token's history through an RPC, a thousand signatures at a time, and decode each one, and for one token that works. For every token, live, it does not. That is the difference between raw access and analytics.
The common metrics, and how each misleads
| Metric | How it is computed | How it misleads |
|---|---|---|
| Holder count | Token accounts with a balance above zero, grouped by owner | Counts wallets, not people. A thousand holders can be one person with a script. |
| Top-10 share | The ten largest owners' balances over supply | Meaningless unless pools, curves and burn addresses are excluded first |
| Volume | Sum of trade sizes over a period | Wallets trading with themselves create any volume they like for the cost of fees |
| Liquidity | The reserves in a token's pools | Can be withdrawn by whoever added it, unless it is locked or burned |
| Market cap | Last price times supply | A price from a nearly empty pool produces an enormous, unsellable number |
| Buys against sells | Count or volume of each over a period | Many small buys and one large sell read as bullish by count |
| Wallet profit | SOL out minus SOL in, per token, for a wallet | See the next section; this is the one most often wrong |
| Early buyers | Wallets that bought within seconds or blocks of creation | Only knowable if the indexer saw the token's very first trade |
Wallet profit, done properly
Three numbers are routinely blurred into one.
- Realized: SOL received from sales minus SOL spent on the tokens that were sold. This has actually happened.
- Open: the tokens still held, valued at the current price. This is a claim, not a result.
- Sellable: what those held tokens would really fetch if sold now into the pool's actual depth. On a thin pool this can be a small fraction of open.
Then the complications. Tokens transferred in from another wallet have no purchase price, so profit on them is undefined unless you follow the transfer back. Fees, tips and account rent belong in cost. A wallet that holds a token through a migration from a launchpad curve to a pool needs both venues decoded or its sale will look like a purchase from nowhere. A leaderboard that ignores these ranks the luckiest accounting artefacts first. How to read a Solana transaction shows the netting for a single swap.
Wallet behaviour analysis
Beyond totals, the useful questions about a wallet are about pattern.
| Question | What is measured | Why it matters |
|---|---|---|
| Is it early, or just fast? | Entry time relative to creation, across many tokens | Early on a few winners suggests judgement. Early on everything is a bot. |
| Does it take profit? | Share of positions closed in profit against held to zero | A record of paper gains that were never sold is not a record |
| Who funded it? | The source of its first SOL | Wallets funded from one source in sequence are usually one operator |
| Does it act alone? | Other wallets that buy the same tokens in the same seconds | Co-ordinated groups look like broad demand on a holder list |
| Is it the creator? | Links from the token's deployer by funding or transfers | "Smart money" that is early on its own launches is an insider |
None of this is identity. A label such as smart money, sniper or insider is a conclusion drawn from behaviour, and it can be wrong. The honest form of every such label is "this wallet behaved like one".
Snipers and bundles
Two launch patterns are worth knowing because they are only visible on chain. A sniper is a wallet that buys in the first blocks after a token is created, usually by software. A bundle is a group of buys placed in the same block as the creation, typically by the creator through several wallets, so that a large share of supply is held by one party while looking spread out. Both are measured the same way: trades whose slot is at or just after the creation slot, and what share of supply those wallets still hold. Both need the indexer to have seen the token from its first transaction; a token picked up later can only be marked unknown, not clean.
What on-chain data cannot show
- Who a wallet belongs to. Only what it did.
- Trades inside centralized exchanges. Those happen in the exchange's own database; the chain sees only deposits and withdrawals.
- Intent. A transfer between two wallets could be a sale, a gift or one person moving funds.
- Anything off chain. Agreements, promotion, who is talking to whom.
- The future. Every metric here describes what has happened.
Building a small one yourself
- Pick one program and subscribe to its logs over a WebSocket. The WebSocket tutorial has working code.
- Decode its trade event into a row: time, slot, token, wallet, side, SOL amount, token amount.
- Store rows in a database with an index on token and one on wallet. SQLite is enough to start.
- Keep a running balance per wallet per token as rows arrive. That is your holder table.
- Compute metrics as queries: top holders, first buyers, each wallet's SOL in and out.
- Handle the gaps: reconnects, backfill, tokens you started watching late. Mark what you did not see as unknown rather than zero.
One program is a weekend. Every launchpad and pool a token can move through, kept correct through program upgrades, is the hard part.
Where AI fits
A language model cannot read a blockchain; it has no access to live data and will invent numbers if asked. What it can do is read numbers that an indexer has already produced, and explain or compare them. The standard way to connect the two is a tool interface such as the Model Context Protocol, where the model calls functions that return real data. How to use MCP for crypto shows the setup, and the BLIX MCP page lists the 26 tools an account's connector exposes.
BLIX as an example of the pipeline
BLIX is an on-chain analytics system with a trading screen on top, built for Solana memecoins specifically. It decodes launches and trades as they happen, keeps a holder ledger per coin with creator, sniper and bundle shares, computes realized, open and sellable results per wallet, and groups wallets linked by funding. The same tables are available to an account as JSON through the API. It does not cover validators, NFTs or programs outside token trading, and where it did not see a token's first trade it says so. Questions about how a number is computed are welcome in the Discord.
The BLIX community
Every account comes with the BLIX Discord: the people trading on the terminal, the narratives as they form, feature drops before they are announced, and the team answering questions. Join the Discord once you have signed up.
More on Solana and on-chain data
This page is education about how the technology works. It is not financial advice, and nothing here recommends buying or selling anything.
Frequently asked questions
What is on-chain analytics?
Analysis of data recorded on a blockchain itself: transactions, balances and program activity. It answers questions such as who holds a token, who bought first and what a wallet has made, using the public record rather than exchange data.
What is blockchain indexing?
Reading every transaction as it happens, decoding it, and storing the result in a database arranged for queries. Without an index, questions that span many transactions cannot be answered quickly.
How is wallet PnL calculated?
Realized profit is SOL received from sales minus the cost of the tokens sold. Open profit values what is still held at the current price. Good tools keep them separate, and also show what the held tokens could actually be sold for.
Why do holder counts differ between sites?
They count different things: empty token accounts or not, pools and curves or not, accounts or owners, and data of different ages. Token accounts explained covers the detail.
Can AI analyze blockchain data?
Only when it is given the data. A model connected through a tool interface to an indexer can read and explain real numbers. A model asked from memory will guess.
Is on-chain analysis enough to judge a token?
No. It shows who holds and trades it and how the supply is arranged. It does not show intent, identity or what happens next.

