WHITEPAPER · V0.1 · NOV 2026
Colony Zero
What this is
Colony Zero is an ant colony that trades memecoins. The ants are simulated and the trades are on paper.
Three hundred of them wander across new tokens on Robinhood Chain, Solana and BNB Chain. When an ant finds a token that lots of different wallets are buying, it carries the news home and leaves a scent trail behind it. Other ants follow strong trails. Weak ones fade. Once a trail gets strong enough, the colony buys the token on paper, and it sells when the trail dies down.
No language model makes these calls. We do use one to write the daily journal, because ants can’t write. It describes what happened. It doesn’t get a vote.
Everything is public: the rules, every entry and exit, and the running score.
Why we built it
Most AI agent tokens are a chatbot with a ticker. Ask one why it bought something and you’ll get a different answer every time. We wanted the opposite: a simple rule you can watch work, or watch fail.
Ants are a good fit for that. A single ant knows almost nothing. A colony still finds the best food nearby, drops sources that run dry, and ignores most false leads, because one ant’s discovery only matters once other ants confirm it. That last part is exactly what launchpads need, where a lot of the volume is wash trading and wallets buying from themselves.
How the colony decides
One tick is 30 seconds. On every tick:
- Each ant picks a token to visit. It leans toward tokens with a strong trail and good food.
- Food is mostly a count of different wallets buying. Net buying in dollars, liquidity and token age add a little. A token loses food when one whale does most of the buying, when the buyers are wallets created minutes ago, when money loops back to the deployer, or when the deployer sells half or more of what it bought.
- Ants that found food lay scent on the way home.
- All scent evaporates by 5% per tick. A five-minute pump is gone before it matters, while a token people keep buying all day keeps its trail.
- The colony enters when a trail crosses a threshold and enough ants are on it. It exits when the trail drops below half of its peak.
For anyone curious, this is standard Ant Colony Optimization:
τ is scent, η is food, ρ is how fast scent evaporates, and α and β set how much the ants care about each.
Left alone, that rule snowballs: the first thick trail pulls in every ant and the colony stops looking around. So when an ant picks where to go, scent counts for no less than 0.3 and no more than 2.5 (the Max-Min Ant System), and one ant in ten ignores scent entirely and goes by food, like a scout. The real trail still decides when the colony buys and sells.
| Setting | Starting value |
|---|---|
| Ants | 300 |
| Tick | 30 seconds |
| Evaporation | 5% per tick |
| α, β | 1, 2 |
| Scent cap when choosing | Between 0.3 and 2.5, so one trail can't take every ant |
| Scouts | 1 ant in 10 ignores scent and goes by food alone |
| Entry | Trail of at least 1.0 and at least 8% of the ants on it |
| Exit | Trail below 50% of its peak, or the deployer sells half or more of its tokens |
| Positions | Up to 5 at once, 20% of what the book is worth now, never more than 5% of the pool |
| Starting book | $10,000 per generation; below $100 the generation starves and a new one starts |
| On the map | 5 tokens per habitat |
| Map bar | 20 minutes old, $500 liquidity, 15 buyers in 15 minutes; on Solana and BNB, no older than three days |
These are first guesses. We’ll tune them on real data and log every change after that.
Where the data comes from
Thousands of tokens launch every day and the colony ignores most of them. A token only shows up on the map after it’s at least 20 minutes old, holds at least $500 of liquidity, and has had at least 15 different wallets buy in the last 15 minutes. Pons tokens start small, so the bar is set where real ones actually are. Even then, only the five strongest tokens per chain are on the map at once. A newcomer only pushes out the weakest if it’s clearly better.
| Habitat | New tokens from | Data |
|---|---|---|
| Robinhood Chain | Pons, then its Uniswap v4 pools | Launches, bonding-curve trades and v4 pool swaps, read straight from the chain every 30 seconds. Pools paired with ETH or a dollar stablecoin only, until we can price the rest. |
| Solana | pump.fun, PumpSwap, Raydium | GeckoTerminal pool summaries: buyers, volume, liquidity and age over fifteen minutes |
| BNB Chain | four.meme, PancakeSwap | GeckoTerminal pool summaries, as on Solana |
On Solana and BNB Chain we only see pool summaries, not single trades, so the whale, fresh-wallet and deployer checks can’t run there yet. The colony still counts different buyers, which does most of the work, but those two habitats are easier to fool than Robinhood Chain until we add per-trade data.
The glow behind each habitat on the site shows how much of the colony is foraging there, so you can see at a glance where it’s moving.
Paper trading
The colony never touches real money. There’s no wallet, no keys and no orders. Each trade is priced from the token’s live price at that moment, and then we take off fees, estimated gas and estimated slippage. If a trade only looks good before costs, it loses here too.
Paper fills have limits. A token that dumps in the seconds before a real order would land, or a pool too thin to fill at all, can make real results worse than ours. Read the numbers as what the rules would have done, not as what money would have made.
Each colony starts with $10,000 of paper money and sizes every trade from what it has left, so a losing streak makes it smaller, the way it would for a person. There are no top-ups. If a colony’s book falls below $100 it starves: whatever it still holds is sold, the generation goes into the record with how long it lived, and a new generation leaves the nest with $10,000 and no scent. The control colonies live and die by the same rule.
The narrator
A language model reads the colony’s state and writes a journal entry every day. When something big happens, like most of the colony moving to another chain, it drafts a post for X, and we post it by hand. You can also ask it why the ants are on a token, and it answers from the trail and buying data.
It can’t pick tokens, change settings or guess prices. The numbers in its writing are filled in from the data, so it can’t misquote them.
People will try to game it
If enough people watch the colony, someone will try to lead it: wash-buy a token, wait for the ants, then sell to anyone copying. Here’s how we make that hard.
- Food counts different wallets. A thousand trades from ten wallets is barely a snack.
- Fresh wallets and money that loops back to the deployer count for very little.
- A trail needs ants coming back with food over several ticks before the colony enters. A single burst evaporates first.
- Live positions appear on the site after a delay, so nobody can trade ahead of the colony.
- There’s no copy-trade button, and there won’t be one.
Out in the open
We publish the rules before the results: the settings, the food formula, every entry and exit with the trail strength that triggered it, and the full score history. When we change a setting, we say what changed, when, and why, and we report results from before and after separately.
The colony also trades beside two control groups under the same rules, ants that ignore scent and ants that wander at random, plus a plain hold of the stablecoin. If the random ants beat ours, you’ll see it.
The token
The token launches on Pons, on Robinhood Chain. It’s how people follow and back the experiment. It doesn’t give anyone a share of what the colony makes, and the colony only trades on paper anyway. Creator fees from Pons pay to keep it running: data, hosting and the narrator.
Roadmap
All three habitats are open. Robinhood Chain gets the full set of checks; the other two catch up in phase three.
Lab
Backtest on Robinhood Chain data, tune the settings, publish the rules.
Launch
The colony goes live on Robinhood Chain. Journal starts. The token on Pons.
Migration
Per-trade data for Solana and BNB Chain, so they get the same checks as Robinhood Chain. The migration log goes live.
There are no dates yet. We’ll set them once the colony has run long enough to judge.
Disclaimer
Colony Zero is an experiment. It makes no real trades, holds nobody’s money and isn’t an investment product. Paper results are simulated and can differ from what would happen in a real market. Nothing the colony, the narrator or our accounts post is financial advice. The token gives no share of profits and no claim on any result.