A reading nobody took
is never a reading of zero.

Five readings, weighted, and then renormalised over the ones that could actually be taken. A token where only three of the five are measurable is scored out of those three, and the share that was measured is reported beside the number as coverage. A pillar with no inputs returns nothing at all. It never returns a neutral fifty.

A critical finding clamps the total instead of being averaged into it. A honeypot with locked liquidity, fair distribution and good momentum would otherwise score in the sixties, and it is still a honeypot.

Splitting a bag across twelve wallets defeats top ten concentration, deployer percentage, holder count and every other distribution metric there is, and it costs one transaction per wallet. What it cannot hide is that all twelve were funded from one place.

Bundle

Supply taken in the token’s first traded block. Not the deploy block: blocks here are about 102ms and the deploy block is empty of buys on all but one token in seventeen thousand, so a deploy block definition reports 0% for the entire chain.

Snipers

Supply taken in the first fifteen seconds. Measured rather than chosen: 31% of all buyers on this chain arrive inside that window, which is why it is weighted far below the bundle. A signal that fires on a third of the population is not a signal.

Clusters

Each buyer’s first inbound funding transfer, chased through the index. A funder seen behind more than forty unrelated wallets is demoted to a hub and stops identifying anyone, so exchange withdrawals do not flag every popular token.

A launch is measured once and it has to be measured whole. If we were not watching when the window opened, the reading stays empty rather than reporting a clean 0%. That is the difference between nobody bought and nobody was looking.

Two sources, in this order.

What already happened

Three or more unrelated wallets having sold in the last six hours is a record, not a simulation, and no honeypot can fake it.

What would happen

For a token too new to have any, a sell of a real holder’s real balance is simulated against the real router, down a ladder of sizes from a quarter of the balance to a thousandth. The ladder exists because a quarter of a real bag is a 19% price move on this chain’s pools, and the check was therefore failing on exactly the tokens most likely to be a honeypot.

Neither answering leaves the result unknown, and unknown is not a pass. The auto trade engine refuses a token whose sell could not be confirmed.

A score is a continuous opinion that changes every minute, so it is never wrong. A call is a moment: at 12:05, at a $20.8K market cap, this was worth buying. It can be measured afterwards, and it can be wrong.

The grade is a floor, not the trigger. The first version of this fired on the score alone and it was both slow and wrong more often than right. Backtested with no look-ahead over 3,939 observations of 781 tokens, the score turned out to predict steeply but only at the very top: launches scoring 80 to 89 reached 1.3x within half an hour 47% of the time and those above 90 did so 80% of the time, while everything between 50 and 80 sat between 9% and 15%. The threshold had been set at 68, inside the band where the score says almost nothing.

What a call has to clear now, with what each gate is worth.

Depth

The strongest single filter. Under $3,000 of liquidity 26% of candidates reached 1.3x; between $3K and $10K, 67%; above $10K, 83%. The three tiers are these three bands: a watch starts at $1,000, a buy at $3,000, a strong buy at $10,000 with twenty or more buyers behind it.

People, not fills

At least ten distinct wallets bought it in the last ten minutes. One wallet buying forty times is one opinion bought forty times; ten wallets is ten decisions. Fewer than five buyers reached 1.3x 5% of the time, forty or more did so 41% of the time.

A two-sided market

Buys must be between 55% and 85% of the flow. Both ends matter and the upper one is the surprise: a token where 100% of trades are buys reached 1.3x only 7% of the time, because that is the shape of something nobody has sold yet. Somebody choosing to hold against other people choosing to leave is the thing worth following.

Which also replaces the honeypot check

Those sells are not a simulation of an exit, they are a record of several, in the last ten minutes, by strangers. No honeypot produces that, and it is strictly stronger evidence than the simulation it replaced.

Age between three and forty-five minutes

Under three minutes the price is whatever the first few fills made it. After an hour the edge is gone: tokens older than that reached 1.3x 7.5% of the time against a 21% base rate.

Near its own high, but not at it

Within 18% of the highest price it has traded. Buying the exact top tick did worse (16%) than buying a shallow pullback of 2% to 15% (28%), and a rule that demands a new high buys every blow-off.

The bundle, as a veto

A known bundle above 20% of supply refuses the call outright: those reached 1.3x 3.6% of the time. An unmeasured bundle no longer blocks anything. A moderate one, between 8% and 20%, was in fact the best band of all at 37%, which is not what anyone expected.

Silence

No call at the same tier in the last two hours, and no live call at a stronger one.

Then it is tracked for six hours: entry, peak, where it is now, its deepest drawdown, and how it ended. Six rather than the week it used to be, because the trade is over long before that. Measured across the calls that reached 1.3x at all, the time from the call to its peak was four minutes at the 25th percentile, ten minutes at the median and thirty-five at the 75th.

A call is withdrawn before that if it stops standing: the grade collapses under it, or twenty five minutes pass and the token has not moved ten percent in either direction. The second is the same distribution read the other way. If three quarters of the calls that work have peaked by thirty-five minutes, a call that has done nothing at twenty-five is not early. A withdrawal is not an outcome and it is not a loss: it is the claim being taken back, and a call that went up and came back is never withdrawn, because that call moved.

Withdrawing a call removes it from the feed and never from the arithmetic. The record below counts every call made in its window whatever happened to it, so a withdrawn one is still in the denominator, still in hit 2x, still in fell 50%. A feed that could retire its disappointments would be worth nothing.

What to distrust here. Every component above rests on thousands of observations, but the combined rule fired twelve times in the window it was fitted on. Twelve. Split in time it held, and it beat the rule it replaced on both frequency and hit rate, but no single number on this page should be treated as settled until the same backtest has been run against a week of data.

Anybody can publish calls. What makes them worth reading is that every one of them, including the ones that went to zero, is counted in the same number, and that the number is served from the same rows the feed is served from so the two cannot disagree.

The strip above the calls reports how many hit 2x, how many hit 5x, how many fell by half, how many rugged, and the median peak alongside the median of where those calls are now. The gap between those last two is the honest measure of a feed like this one. A product that publishes only peaks is showing the best moment of every call it ever made.

Medians rather than means throughout. One 400x carries a mean and says nothing about what the typical call did.

DexScreener does cover this chain, which makes it the one external source in the product. It answers two questions no amount of chain reading can: has anybody paid, and how much attention did they buy.

It is worth at most a few points either way, and it is never a pillar. A filled profile is what a real project does and most scams do not bother with, so it helps a little. A community takeover means the original team left, which is a fact about the launch, so it hurts a little. Advertising counts for nothing at all: it is exactly as likely to be a marketing budget as an exit being planned, and it is a purchase rather than a property of the contract.

The adjustment is applied after the five pillars are weighted and before a critical finding clamps the total, so nothing bought can lift a honeypot back over its ceiling. It does not raise coverage either: coverage answers how much of the token could be measured, and a payment is not a measurement of the token.

Prices and liquidity are not taken from there. Two sources for one number is how a product disagrees with itself on the same screen.

The deployer is the wallet that deployed the contract, as the launchpad reported it. A deployer trading from a second address is invisible to the marks on a chart, and that is the honest limit of the claim. Labelling a wallet as the developer on a funding heuristic puts an accusation on somebody else’s chart.

The call thresholds are not yet calibrated against a long run of outcomes. Nothing on the bench is advice, and the engine will not act on a token it could not measure.