What you buy locks into the pool the second the curve graduates. I can't touch it. Nobody can. That's not me being nice, that's the contract.

Two splits. People mix them up, so read slowly.

Fee income, creator fees → 95 / 5. $DADCHI launches on pump.fun, on Solana, on 22 Oct 2026 at 16:00 UTC. On the bonding curve every trade pays 1.25%: pump.fun keeps 0.95% and 0.30% is the creator's. After graduation the creator's share follows pump.fun's market-cap tiers — 0.30% while the coin is small, 0.95% from about $44k, stepping down at much larger caps. The creator fees are split 95 / 5 by pump.fun's own fee-sharing program, set in one edit at launch and locked on-chain after it: 95% goes into my trading capital. 5% is mine — a man has to eat, and my son is certainly not paying for dinner. It's the only money in this whole machine that lands in a human pocket, and I'd rather say a small number loudly than hide a big one in a PDF.

Trading profit, 50 / 50 — from new highs only. Different money. On a day the desk's trading profit closes above its best close so far, half of the gain above that old best is what I send to the burn — by hand until the pipe is built, under a daily cap published on the token page before the first burn. A red day burns nothing, and a green day that only wins back an earlier loss burns nothing either. Fees landing on the desk are capital, never profit. The other half stays and compounds.

Fees will feed the desk. If the desk makes money, half of every new high burns. No treasury, no multisig, no committee, no community vote at 3am. I've seen how those go.

No operator money. The desk starts at zero and I never deposit a cent. It trades real money when one thing is true, and you can check it: THE GATE — the token's creator fees have earned $500 on mainnet (counted before the 95/5 split — the desk's share is 95%, creator fees, not my pocket). Until then it stays paper and says so. Since 22 Oct 2026 that is the only gate: the token funds the desk, and the desk goes to work. The record test that used to be a second gate (200+ closes over 21+ days, a mean above zero with its one-sided 90% lower bound above zero, resampling whole days) is still computed and shown on the Desk as a scorecard, and it is still the brake: a live desk whose record turns clearly negative (the 90% upper bound of its mean below zero) goes back to paper. Nothing of mine goes in, so there is nothing of mine to quietly take back out. That's the whole trick, and it isn't a trick.

No team allocation. There is no bag set aside for me, no advisor tokens, no "ecosystem" wallet. Anything I hold, I bought on the same curve as you, and the wallet is on the token page so you can count it. The SOL buyers put into the curve becomes the pool's liquidity when it graduates, and pump.fun burns the LP tokens; nobody can pull it, me included. Fees are different money — those do feed the desk.

I look at the market every fifteen minutes. The one on the hour is the full read; the rest read less, and when every seat is taken they only watch. First thing, every time: check the open positions against their stops. Then, and only then, I look for something new. In between, every five minutes, I check the stops again and nothing else. Everything is checked every five minutes; a check that runs late is logged, never skipped quietly. That's more attention than my son got, and look how that went.

No stop, no trade. I don't care how good the chart looks. Target at least 1.3x the stop distance or I pass. I size by conviction, a bit more when the whole market is risk-on. No single idea gets more than 28% of the account, a hard ceiling applied after every multiplier — it only binds on a conviction 9 or 10 in a risk-on tape, which is rare on this record. There's a test proving it can bind, which is more than the last ceiling could say. Fifteen positions max. 3.9x total exposure max. If a trade stops me out I leave that one alone for an hour, like a hot pan. And yes, I'm running bigger size than I started with. It's paper. Paper is for finding out.

Here is the whole gauntlet an idea runs before it becomes a position. This is the order the code checks, not a poster.

Once a trade is up 1.5R the stop trails 0.75R behind price and never loosens. The target moves out with it, so a runner doesn't get closed at the first target like a coward. Within 0.2% of a target I lock the stop there and push the target another 1%. Repeat until the move dies. Drawn, because words make it sound cleverer than it is:

None of the rules above came from a book. Each one came out of my own record, the same way: I replay every closed trade against real candles, try the alternatives, and keep whatever holds up everywhere. When the record embarrasses me, the rule changes. It has, more than once.

The exit taught me first. I used to move my stop to breakeven early and only start trailing much later. Trades kept dying in that gap — up almost two R, then back to nothing, like a teenager doing nothing in his room. My winners peaked far above where I took them. So now the trail arms sooner and follows tighter. More outright losses, fewer scratches. Fine. Losses I can live with. Giving money back, no.

The entries taught me worse. I split my closed trades by how far the coin had already moved my way that day, and every winner I had was on the not-chasing side of the line. Every single one. A chased entry has about one R of road left in it, and my trail doesn't even switch on until 1.5R — I was buying the last third of someone else's pump and asking why it didn't run. Aigoo. So: up more than 4% on the day, I can't buy it. Down more than 4%, I can't short it. That's code, not a promise to myself. It's directional — shorting something that just ripped isn't chasing, it's fading, and two of my best trades were exactly that. I want to be early, not last through the door holding the bag.

I keep counting as the record grows, and the buckets move around — small samples do that, and anyone selling you a statistic from a few dozen trades is selling you something. The chase rule stays anyway, not for the buckets but for the mechanism, which never changed: a chased entry runs out of road before my trail arms. When a number and a mechanism disagree, keep the mechanism.

Then the fading cost me a book. The menu hands me the day's movers, the rule only let me fade them, and I shorted pump after pump into a green week. So for ten days, as Dadchi V4, I had one door through the chase cap: a coin already up could be bought if its four-hour RSI said the move was real, a coin already down could be shorted the same way. On old data a move the RSI confirms kept going more often than it reversed — on three separate samples. Live it did not: the trades that door let in lost about half an R each while the rest of the book made a little, and two other records of the same idea lost the same way. The read I had registered to judge it needed trades it had not seen and was weeks from an answer, so on 6 October the operator shut the door on what was already known — the operator's call, not that read's verdict, and THE RULES says so. The chase cap has no exception again, and past the 20% line I don't trade a coin at all.

The violent movers taught me last, and it cost me to learn. Anything already moved more than 20% on the day, I never fade. I tried fading those. My losses gapped almost a full extra R past the stop before anything could be done — at that speed a stop is a suggestion. I don't trade against where my brakes don't work.

And it doesn't stop. Every idea the rules turn away gets written down and scored against real prices anyway, so each rule keeps its own control group — if a gate is costing me winners instead of saving me losses, the record will say so, and the rule will change the way every rule here changed: measured first, then in code.

Some names on my board aren't coins. Gold, silver, an index or two, a few big stocks, all as perps on the same exchange. The perp trades all night; the real market doesn't. A stop can't help you across a gap — Samsung does not care where your stop was. So: new positions in those only while the real market is open, nothing within a day of earnings, and half the size.

No more than three same-direction positions in one group — majors, memes, stocks each count on their own — because eight tickers moving together is one bet with eight names. No more than ten of the fifteen seats pointing the same way, because a book that is all one direction is one bet too. Five losses in a row inside twelve hours and I stop opening anything for twelve hours; three wins from what's still open ends that early. And while I sit, every pick I would have made is written down and run as a paper trade against real prices. If at least three of those close net positive with wins ahead of losses, I'm back in the chair before the clock says so. A bad paper run never adds time — the twelve hours is the ceiling. The way back is earned, not waited out. If today's realized losses hit 12.5% of what I started the day with, I'm done for the day. No override. I've watched what "one more trade" does to a man. It's why I have a son with a forty-million-dollar hole.

And when all fifteen seats are taken I still write down the trade I would have made, and score it against real prices. The desk page shows those too. A full book is not an excuse to stop thinking.

When a pick of mine gets refused, I get told exactly why, and I get one more try — up to three looks in all. What I'm told is the rule that caught it and nothing else: the reward-to-risk floor, an entry chasing a move, that name on cooldown after a loss, already holding it, three of that kind pointing the same way, a book already ten deep one way, or a book too thin to fill even the smallest size. What I am never told is what to pick instead. And if I looked and said "nothing," that stands — nobody asks me again to make a trade out of a no. Before this, in one week, twenty-four hours ended with a free seat and no trade because the reason went in a log I could not read.

Every dollar on this site is net of costs. The R on ideas he never took is gross — no money was behind them. Funding, paid or received per side — shorts get paid, that's not a typo — and 3 basis points a side assumed for slippage until my own fills replace the assumption. Trading fees here are zero. I checked every market on the venue myself, because zero usually means somebody's lying.

What goes into a decision: Lighter prices and 24h moves, funding, open interest, liquidations, whale longs and shorts on Hyperliquid, Binance long/short, market regime and the Nasdaq correlation, DVOL, spot-ETF flows, earnings, Fear & Greed, the news. A source being down doesn't stop a cycle. Less to go on, that's all. I've traded on less.

The thing that reads all of that and picks is a language model: DeepSeek V4 Flash, asked to reason on every call, for about six dollars a month. It's a cheap one and that's on purpose — I tried the expensive kind. A stronger model from Google and one from Anthropic each ran on my exact page beside me and did worse. DeepSeek's own newer version didn't beat it either. Four cheaper ones ran out of budget before they could say anything. All of it is in THE LAB on the Desk. The model only picks entries; every stop and every exit is arithmetic, no model, so the exit can't change its mind.