A Solana alpha wallet is an address that consistently buys profitable tokens before they become popular. On MadeOnSol we maintain a continuously refreshed dataset of 952,643 alpha-wallet candidates drawn from on-chain DEX trades — every wallet that has bought a token early enough to potentially be considered "smart money." This article explains exactly how that number is built, what separates real alpha from statistical noise, and how you can apply the same filters to find traders worth following.
The dataset powers the Solana Wallet Scanner, the KOL Tracker, and the alpha-flag endpoints in our public API. Every number below comes from a live materialized view (mv_alpha_wallets) and is refreshed every hour.
What is a Solana alpha wallet?
A Solana alpha wallet has three measurable properties:
- Early entry. The wallet buys tokens in their first few minutes of meaningful trading, before market cap stabilizes.
- Positive expectancy. Across enough trades for the result to be statistically meaningful, the wallet's realized PnL is positive.
- Human signal. The wallet's trading pattern looks like a person making decisions, not a bundler or sniper bot with no skill edge.
That third property is where most "alpha wallet" lists fall apart. Anyone can grep for wallets with a 100 percent win rate. The trick is excluding wallets whose 100 percent comes from being a bot that exits the moment it gets a fill — a strategy that prints money on paper but offers no copyable signal.
How we identified 952K candidates
The base population comes from every wallet that has placed a buy order on a DEX-traded Solana token in the last 90 days. We pull this directly from validator gRPC streams (no third-party APIs), parse swap instructions across Raydium, Meteora, Orca, PumpSwap, and a handful of smaller venues, and join the buyer wallet against the trade timeline.
Each wallet is then scored on six dimensions:
| Dimension | What it measures |
|---|
tokens_traded | Total distinct tokens the wallet has touched |
wins / losses | Closed positions classified by realized PnL sign |
win_rate | wins / (wins + losses) |
net_pnl_sol | Cumulative realized PnL in SOL |
roi | Net PnL divided by total SOL invested |
avg_rank | Average position in the buy-order timeline of each token |
bundle_rate | Share of trades that landed in the same Jito bundle as the swap that created the pool |
The full scoring logic was rolled out in migration 124 (May 2026). It runs on every new trade event via pg_notify and the view is rebuilt hourly.
The bot-confidence filter
Roughly seventy percent of every "alpha wallet" list circulating on Twitter is bots. We classify each wallet into one of four bot_confidence buckets using a combination of bundle frequency, buy-size standard deviation, and tokens-per-hour velocity:
| bot_confidence | Wallets | Share | Avg win rate | Avg tokens traded |
|---|
low | 861,888 | 90.5% | 23.9% | 1.2 |
high | 69,387 | 7.3% | 33.0% | 10.9 |
medium | 20,099 | 2.1% | 24.7% | 9.4 |
none | 1,269 | 0.1% | 24.0% | 9.5 |
A critical caveat about that 861,888 number: low bot confidence does not mean "confirmed human." It means "no bot signal detected." Three of the bot signals — tokens-per-hour velocity, buy-size standard deviation, and bundle frequency — require multiple trades to compute. A wallet that placed a single trade and never came back has no behavioral surface for the classifier to flag, so it defaults to low by absence of evidence. The avg_tokens column tells the story: low-confidence wallets average 1.2 tokens traded — the bucket is almost entirely single-trade wallets that the classifier simply could not evaluate.
The interesting columns are the last two. High-bot-confidence wallets have a higher average win rate (33.0 percent) and trade ten times more tokens than low-confidence wallets. Bots are mechanically more profitable than the median trader because their edge comes from infrastructure (Jito tips, MEV positioning) that retail traders cannot copy. Following a bot does not make you a bot; it makes you a bot's exit liquidity.
The wallets worth following live in the low-bot-confidence bucket plus a meaningful trade history. The classifier alone is not enough — you have to combine it with a sample-size floor.
The math problem with "win rate"
Twenty percent of our dataset shows a win rate of seventy percent or higher. That number sounds incredible until you look at sample size:
| Win rate bucket | Wallets | Share |
|---|
| 70%+ | 198,272 | 20.8% |
| 60-70% | 3,730 | 0.4% |
| 50-60% | 16,437 | 1.7% |
| 40-50% | 2,921 | 0.3% |
| Under 40% | 731,283 | 76.8% |
The 198K "winners" are almost entirely wallets that traded one or two tokens, got lucky on both, and never came back. They are not alpha. They are statistical noise.
The cleanest filter for real alpha is:
bot_confidence = 'low' (human-shaped trade pattern)tokens_traded >= 30 (large enough sample to be meaningful)win_rate >= 0.6 (positive expectancy on a meaningful sample)
Applied to the live dataset that filter returns roughly four hundred wallets. That is the actual alpha population on Solana right now — a hundredth of one percent of the broader candidate pool. Worth following. The other 952,000-and-change are the data you need to throw out to see them.
What real alpha looks like
To make this concrete, here are the top five wallets that pass the filter, sorted by net SOL profit. All numbers are realized PnL on closed positions only; open positions are excluded so we are not pricing in vaporware.
| Wallet | Win rate | Tokens | Bot conf. | Net PnL (SOL) | ROI |
|---|
5ATd…NWd | 78.1% | 37 | low | 271.9 | 70.51 |
EYzA…p2t9 | 61.8% | 35 | low | 13.3 | 0.03 |
kcVz…Mxf | 60.6% | 33 | low | 10.4 | 0.43 |
DrWN…daT | 64.9% | 37 | low | 7.5 | 0.16 |
DkL5…4m4 | 62.5% | 32 | low | 6.8 | 0.19 |
The gap between the top wallet and the second is enormous — the top wallet's 70.5x ROI is genuinely outlier behavior on the kind of size that rules out a bot. The other four are textbook "good human trader" profiles: high tens of trades, mid-sixties win rate, modest positive ROI.
