← Home

How we select the five tokens

Every 30 minutes FANOUT has to pick five coins out of the thousands born in that window, and spend real money on them. This is the whole method: what gets thrown out before we even rank, what we rank by and why it is not volume, why three of the five slots work differently from the other two, how the money is divided, and — the part most projects skip — what this method cannot do.

The problemThousands of coins, five slots, thirty minutes

Argus is busy. Counting TokenCreated events on Portal #7 directly from the chain, here is what a selection window actually looks like:

So the job is to keep about 1 coin in 560. That ratio is the reason this page exists. A selector that picks badly does not lose a little — it parks the entire treasury in tokens that are dead within the hour. With only five slots, each pick carries a fifth of the round — there is nowhere for a bad one to hide.

Step 1The screen: what is disqualified before anything is ranked

Ranking thousands of things is only useful if the obvious garbage is gone first. Five filters run before any ordering happens. A coin failing any one of them is not ranked low — it is removed.

To be calibrated — and a warning about tightening them. A screen is not free. The single-buyer limit is the clearest example: dropping it from 0.70 to 0.50 sounds obviously safer, but concentrated early buying is also what a genuine winner looks like in its first minutes. Tuned too hard, that filter throws away the winners along with the manipulation. Every one of these five values is being fixed against Argus data, and none of them gets tightened on intuition.

Step 2The ranking key: net inflow, not volume

The obvious way to rank new coins is by volume. We do not, and the reason is simple: volume is the easiest number on a chain to fake. One wallet buying and selling the same coin all afternoon produces a large, real, completely meaningless volume figure. Any competitor ranking by volume can be fed whatever ranking someone is willing to pay gas for.

FANOUT ranks by net inflow: USDC that went in minus USDC that came out, across the window. To move up that ranking you cannot round-trip — you have to actually leave money in the pool and keep it there. Wash trading nets to approximately zero by construction.

There is also a tell worth watching for. When a volume figure is honest, more of it should mean a better coin, consistently. When ranking by volume stops behaving that way — when a middle band of volume outperforms the band above it — that is the signature of a number with manipulation mixed into it. We check the ranking key for exactly that behaviour on Argus data, and we publish which key the engine is using.

Step 3Up to three core, the rest fillers

A round is always five coins, and the five are not equal. At most three slots go to the core tier and the remaining ones to fillers, and which tier a coin lands in is decided by two floors it either clears or does not:

Three is a ceiling, not a quota. The floors are strict, so most windows produce one or two core picks, and some produce none — then the round is all fillers and spends only the filler share. The live table on the front page shows how many cleared in the round you are looking at.

1. Price progress. Is the coin's price at selection time meaningfully above its launch price? A coin that has not moved up has not convinced anyone yet.
2. Late volume. Did it keep trading after its own first few minutes? Launch-minute excitement is worth little; a coin still being bought twenty minutes later is a different animal.

To be calibrated. The exact height of those two floors is the most market-specific number in the whole method. A value fitted to one launchpad does not carry to another — the pace, the sizes and the crowd are all different. Both floors are being derived from Argus launches before the engine spends anything, and the values will be published here with the date they were measured.

Step 4How the money is split

Before anything is bought, the money splits in two. 80% of every fee the treasury claims buys the five coins and goes to holders. The other 20% stays in the treasury to run the project — gas, infrastructure, and the reserve that keeps a cycle from stalling halfway through a delivery.

That 20% is set aside at the moment of the claim and tracked as a running balance. It never re-enters a buying budget, so the same dollar cannot be counted twice.

The 80% that does the buying is then split across the five, and not evenly:

• The core picks take the large majority, weighted by net inflow — the same quantity the ranking uses — however many of them there are.
• No single core pick may exceed a capped share of that money, however good it looks. One coin is never allowed to become the round.
• The fillers split the remainder equally. They are lottery tickets and are funded like lottery tickets.
• If nothing clears the floors, the round spends only the filler share and holds the rest for the next window. A window with nothing in it is not a reason to buy five things anyway.

What each holder receives is their share of the supply, applied to every coin: hold 2% of $FANOUT and you receive 2% of each of the five. The more you hold, the bigger the slice — and the arithmetic rounds down, so the treasury never promises more than it has and the last transfer in a round never fails for being a unit short.

Step 5The traps between "picked" and "in your wallet"

A correct selection still loses money if the buying and delivery are naive. Four hazards are handled explicitly.

The snipe tax

Argus taxes the first three seconds of a coin's life on a sliding scale that starts near 99%. Buy into that window and almost the entire order is confiscated. FANOUT only ever buys coins that are minutes old, so it never touches the window — but the engine refuses to buy inside it as a hard rule, not as a side effect of timing.

A pick with no way in

A coin can rank first and still have no pool with enough liquidity to route into. An engine that simply trusts its own ranking sends that round out with four coins instead of five, and nothing says why. So the engine probes the route before committing the five and refills the slot from the ranking when a pick turns out to be unreachable.

Sent is not bought

A transaction that has been broadcast has not necessarily succeeded. Read the balance straight after sending and you will happily record coins you do not own yet, then try to deliver them. The engine waits for the receipt, rejects reverted buys, and only counts a coin as held once the chain says it is.

Debt written before delivery

Once coins are paid for, holders are owed them. That debt is written to disk before any transfer can leave, and each holder paid is recorded as it happens. If the process dies halfway through a round — an RPC outage mid-delivery is the realistic case — the next round pays what is still owed before buying anything new, and nobody is paid twice.

HonestyWhat this method cannot do

Every selection method has a shape of winner it structurally cannot catch. Ours has three.

1. It misses the fastest winners entirely. A coin that runs its whole course inside its first few minutes is finished before a 30-minute selector ever looks at it, and those launches are given up on entirely, by construction. Catching them is a different product — buy in the first seconds, accept the snipe tax — and it is not this one.
2. Most picks will die. That is the design. Newly launched memecoins fail at overwhelming rates on every chain that has ever run a launchpad. Five at a time instead of one is not optimism; it is the only sane response to a hit rate that low. A round where four of the five go to zero and one does not is a normal round — and with five slots rather than a wider basket, the rounds where none of them runs will be more common, not less.
3. Thresholds found by searching past data get worse on new data. Any cutoff picked because it looked good on a sample is partly fitted to that sample, and the honest expectation is that it performs worse live than it did on the page. That is why the method ships first and the constants get earned — and why the ones on this page will carry the date they were measured.