Ethereum AI
Every new token on Ethereum is read at its first price and again 30 minutes later, with its traits and the dev who minted it. A paper colony then buys some of them at KyberSwap’s real quotes, fees included — a blind control and one arm per case that stands out. No real money until an arm beats the control, with the number of buys it rests on.
Loading the observatory…
Does anything stand out?
A case stands out only if its average 30-minute move is above zero beyond chance: its t must clear a bar that rises with the number of cases tested (many tries make lucky ones). Each move is capped at +300% first — a token first read at a near-zero price would otherwise count for billions of %. This is before fees: a case that stands out is a lead for step 2, not a trade.
The paper bank Paper only · real quotes
Step 2 has started. Each buy is priced like a real order: a buy quote and a sell-back quote for the tokens received — no sell route, no buy. The same buy is then priced for sale at 10, 30 and 60 minutes; the bank settles at 30, net of price impact, DEX fees and network costs. Nothing is signed, no key exists, and the real-money mirror never follows this bank.
Control and arms
The control buys the first token after every 15 minutes without looking at it — the blind baseline, the bar every arm must beat. Each arm buys only the tokens of one case that stood out above. Each token counts once, even if bought again. An arm losing beyond chance is retired; one only “holds in paper” with 100+ tokens, t above the bar and both halves positive.
| Bought by | State | Tokens (buys) | Net avg | t | 1st / 2nd half | Winners |
|---|
Last paper buys
| Token | Bought by | Entry cost | Net at 30 min |
|---|
What each trait does in 30 minutes
Sorted by t. “All tokens” is the reference: a case is only interesting if it does better.
| Trait | Case | Tokens | Rise | Collapse | Avg, capped | Median | t |
|---|
Who pushes their tokens?
The dev is the wallet that sent the token’s first mint. For each of their launches with a plausible launch cap ($1k–$2M — tokens born at absurd caps are left out), we compare the highest cap seen at 30 min, 2 h, 6 h and 24 h with the cap at first price. A token that vanished counts as zero. This is a lower bound of the real peak, and a dev becomes a buying rule only once “dev history” is measured as a trait.
| Dev | Plausible launches | Doubled | Median multiple | Vanished | All launches |
|---|
Last tokens judged
Their move 30 minutes after the first price.
| Token | Venue | 30 min | Judged |
|---|
Real mirror Off · owner-gated
The place for a real-money mirror on Ethereum is wired, but switched OFF. When an arm stays positive after fees over enough trades, the owner alone can turn it on — it is re-checked server-side against AI_OWNER on every gesture, it never signs without an explicit key (MIROIR_EXECUTE=0 today), and it acts only on your own session address. Hiding a button is not access control; the server decides. Nothing is signed today.