Smart Money Tracking
Every 15 minutes the indexer ranks wallets by their realized profit on Robinhood Chain tokens over the last 14 days, keeps the best 150, and turns what they are doing into four filters any token agent can use.
AIKOLs builds its smart-money list from its own indexed trades, not from a third-party label service. The work lives in the indexer's smart-money tool, and the wallet-level evidence layer that reads the same trade data lives in src/lib/agents/wallet-*.ts.
Who counts as smart money
Raw trades are kept for 48 hours, so as they come in the indexer folds each wallet's buying and selling of each token, per day, into a 14-day record, valued in USD at the quote asset's price at the time. A wallet is the transaction sender, so a trader using an aggregator is still one wallet. Every 15 minutes it rebuilds the list from that record. Profit is realized only: what a wallet got from selling, minus the average cost of the tokens it sold. Tokens it still holds count for nothing, and tokens it sold beyond what it bought (received by transfer) have no cost basis. A wallet makes the list when all of these hold:
| Test | Threshold |
|---|---|
| Tokens traded | At least 3, and at most 150 a day (more is a router or bot, not a trader) |
| Money put in | At least $300 in total |
| Overall result | Realized profit above zero |
| Hit rate | Up $10 or more on at least half of the tokens it sold |
| Persistence | Up $10 or more on at least 3 tokens, or on 2 tokens on different days: one lucky pump doesn't qualify |
| Not an insider | Never deployed or launched a token, was never exempted from a token's snipe tax or paid its creator fees, and never bundled a launch (bought in a launch's first 3 blocks alongside 2 or more other wallets) |
| Rank | Top 150 by realized profit |
The insider test is what keeps a dev cluster pumping its own token off the list: when it was added, 107 of the 150 wallets on the old 48-hour list were deployers, dev wallets, creator-exempt wallets or fee recipients.
What it adds to each token
Once the list is built, each actively traded token gets four values:
| Parameter | Meaning |
|---|---|
smart_buyers_1h | How many smart wallets bought it in the last hour |
smart_net_usd_1h | What smart wallets bought minus what they sold in the last hour, in USD; negative means they are selling |
smart_holders | How many smart wallets still hold it (kept more than 5% of what they bought in the 14 days) |
smart_early_buyers | How many smart wallets were among its first buyers (first 15 minutes); fixed once the token is 2 days old |
These are ordinary filter parameters. Use them in the Custom Strategy Builder, for example smart_buyers_1h >= 2 and smart_net_usd_1h > 0, or rank by smart_net_usd_1h. Using any of them shows the Smart-money wallets tool as connected on the agent's profile. The tool is free, works with token-filter agents (and engine agents' meme trades), and is recorded in the 5-minute snapshots, so it becomes backtestable as history builds.
Wallet evidence (in shadow)
A second, separate layer reads the wallet-level facts behind a single token from the indexer and turns them into evidence about who is behind it:
- Creator: launches, graduations, graduation rate, and the share of supply the creator still holds.
- Concentration: the top buyer's share of all buying, the busiest router's share of trades, distinct buyers and makers.
- Sniping: buys that paid the anti-snipe tax and the share of early buying by tax-exempt (insider) wallets.
- Pressure: buys, sells, buyers, sellers, and net buy ratio ((buy volume minus sell volume) divided by their sum).
A transparent baseline turns that evidence into avoid, caution or ok:
| Flag | Threshold | Effect |
|---|---|---|
| Serial launcher | At least 3 launches with a graduation rate of 10% or less | Avoid |
| Creator holds a lot | Over 50% of supply | Avoid |
| One dominant buyer | Top buyer over 40% of buys | Avoid |
| Insider-heavy launch | Exempt wallets over 50% of early buying | Avoid |
| Creator holds some | Over 30% and up to 50% | Caution |
| One router dominates | Over 60% of trades | Caution |
| Net selling | Net buy ratio below -0.3 | Caution |
An optional reasoning step can ask Claude to weigh the same facts through a written guide to launch dynamics instead of fixed thresholds. Missing data never raises a flag.
The wallet gate (wired in as a signal gate whose only source today is this wallet evidence) does not affect live trades. It runs only in shadow mode (off by default), logging the buys it would have vetoed so they can be scored against what the tokens did next. It would act live only after that scoring shows its vetoes avoid more losses than they cost.
Limits
- 14 days, building up. The record starts when this version went live with the 48 hours of trades then kept, and reaches 14 days two weeks later. Past profit is not a forecast.
- Only Robinhood Chain. Profits on other chains are invisible.
- USD at the time of trading. Each day's buying and selling is valued at the quote asset's price when it was recorded, not at today's.
- Sybil splitting. A trader spreading across many wallets dilutes each wallet's record.
Related
AIKOLs is on Robinhood Chain testnet. Testnet performance is not evidence of real-money profitability.