Paper
Why a token would hand its fees to a model one trade at a time, and what that is supposed to do for the people holding it.
Fees are usually wasted
Most tokens collect a fee on every swap and do one fixed thing with it. Some burn it. Some send it to a wallet nobody can see into. Some pay it back to holders on a timer. A fixed rule is simple, and it is wrong most of the time, because what a market needs changes from one trade to the next. A burn does nothing for a pool so thin that one seller can move price ten percent. Paying holders does little when nobody is selling and the supply is the problem. Adding depth is wasted on a pool that is already deep.
One question per trade
vej does not pick one of those. For every swap it asks a model which of the three the market needs most, given that exact trade, and it lets the model answer with a division instead of a verdict. The fee is cut three ways by the three probabilities. A confident answer sends nearly everything one way. An uncertain answer spreads it. Either way every cent of the fee has somewhere to go the moment the trade is read.
Pressure turned around
The name is the model's name backwards because the machine is meant to push back against whatever just happened. A large sell into a thin pool is the worst thing that can happen to a small token. Here that same sell produces the largest fee, the model reads it as stress, and most of that fee is sent to the pool that was just drained and to the wallets that did not sell. A run of steady buying into a healthy pool reads the other way, and the fee goes to destroying supply. The model is not told to do this. It is shown the facts and asked what is needed. The splits page shows whether it behaves this way in practice, trade by trade.
Weight
The holder share is divided by weight, and weight is balance multiplied by the share of a bag that was never sold. The second half of that is what matters. Two wallets with the same balance do not earn the same. The one that got there by buying and keeping earns in full. The one that sold most of a larger bag and kept the remainder earns a fraction, on every trade, until it buys back. There is no timer on this and no penalty period. The arithmetic is recomputed on every trade from what the wallet has actually done. Splitting a bag into many wallets gains nothing, since each piece carries its own small balance and the pieces add up to the same weight.
A seller pays the people who stayed
When a wallet sells, it is left out of the holder share of its own sale. Everyone else with weight is paid from it. The larger the sale, the larger the fee, and the more likely the model is to lean toward holders. So the wallets that gain most from a dump are the ones that were holding the most, and had sold the least, at the moment it happened.
Why this needed a new kind of model
None of this is practical with a model that writes. Asking a chat model about every trade would be slow, it would cost real money on a busy day, and its answer would be a paragraph. Jev returns a typed answer in a fraction of a second at a price close to nothing, and the answer is already in the shape of a split. It is named for William Stanley Jevons, who noticed in 1865 that making something cheaper to use leads people to use far more of it. A judgment used to be expensive enough that a token could afford one rule for its whole life. It is now cheap enough to make a fresh one for every swap.
What can go wrong
The model can be wrong. It may send fees to the burn in a moment that needed depth. Its answers are stored with the facts it saw, so anyone can go back and judge them, and the question can be rewritten if the pattern is bad. Balances here are built from watched trades, so tokens moved between wallets by transfer are not counted as weight until they are traded. And the amounts on this site are what the rule assigns. They count as paid only when a transfer is recorded.
What to watch
The front page shows the machine running. The raster shows who was paid from each trade. The jar shows what is owed to holders and not yet paid. The splits page shows how the model has leaned. The trades page shows every answer beside the trade that caused it. If the idea works, sells should sit near the holders corner of the triangle, large trades in thin moments should sit near the pool corner, and quiet buying should sit near the burn.
Current figures
| measure | value |
|---|---|
| agent wallet | 7BC451T2s3AmZPLKqcPCpyZcx47EZfCqChks95QjVwhm |
| fees routed | |
| to the burn | |
| to the pool | |
| to holders | |
| trades read | |
| mean holders share on sells | |
| mean burn share on buys | |
| mean pool share on large trades | |
| wallets with weight | |
| wallets that never sold |