24 hours after its peak, the median token is worth 8% of it.

The shape of the fall

Every token here was scored by our scanner before its outcome was known, then polled roughly every 15 minutes for as long as it kept trading. For each one we found the highest price we observed, and asked a simple question: how much of that price is still there later?

Roughly half the peak is gone inside an hour. Two thirds by dinner. By the next day the median token is a rounding error against its own high.

This is measured from each token's own peak, not from launch and not from any entry price. It says nothing about whether you could have bought at the bottom or sold at the top. It describes how long the top lasts.

Our own "low risk" tokens die too

This is the part we would rather not publish, so here it is first. Our scanner sorts tokens into three bands by how much evidence of coordinated control it finds. Twenty-four hours after the peak:

Two things to take from that table, and neither is flattering in the way a marketing page would like.

The bands are not a ladder. ELEVATED holds its value better than LOW SIGNAL. If our score were a clean severity gradient, that would not happen. It is a detector for one specific thing — coordinated wallet control — not a general quality rating, and this is what that looks like in data.

A low score is not a green light. A token we flagged nothing on still keeps about 6.6% of its peak a day later. "We found no evidence of a cabal" is a much narrower statement than "this will hold", and anyone reading the first as the second is going to lose money.

The high-risk band contains the biggest winners

The obvious reading of a risk score is that high risk means the price goes down. That is not what the data says. Splitting by whether a token graduated — because graduating changes everything about how a token trades, and pooling the two produces a reversed result — we get:

In both cohorts the high-risk band is more likely to produce a 10x and more likely to end near zero. That is not a contradiction, it is the same fact twice: a coordinated token is engineered to move. The move is real. The exit is the part that is not there when you need it.

It is also why we no longer print the word "AVOID" as an instruction anywhere a human reads. A community takeover or one well-timed post from someone with an audience can move any of these, and we cannot see either. We report what is on-chain. What you do with it is your decision.

Method, and what is wrong with it

Every number above comes from tokens scored before their outcome was known, so none of this is hindsight. The uncomfortable details:

• We arrive late. Our first observation lands a median 34.9 minutes after the trading pair is created (p25 17.8, p75 61.6, 74.3% within the hour). "Time to peak" and any multiple in this piece are measured from our first sight, not from launch. Anything that happened before we arrived is invisible to us, so our multiples understate the real move.
• Roughly half the time, the peak we see is our first observation. Measured on a random sample of 3,000 tokens: 39.6% of LOW SIGNAL, 52.8% of ELEVATED and 50.8% of HIGH tokens never traded higher than the first price we recorded. For those the real top happened before we arrived, so what we call the peak is a local maximum. This means the decay above is decay from the best price we observed, and the true fall from the true top is steeper than what is printed here — the bias runs against our own headline, not for it.
• 1,435 tokens are excluded for having no usable price series, rather than being counted as total losses. Counting them as deaths would have made every headline here worse and the work less honest.
• Sample: 22,893 tokens first seen between 15 June and 24 August 2026, mean observation window 584 hours. Prices from DexScreener at roughly 15-minute resolution. Tokens with fewer than five observations are dropped.
• The decay figures cannot drift. They are computed from stored price history only. Figures elsewhere that use today's price — the "down 90%" column above — move as the market moves, and are correct as of 24 August 2026.

What this is useful for

Not timing. We are not telling you when to sell, and the spread around every median here is enormous — the tokens that peak in ten minutes and the ones that peak two days later are both in that 8%.

What it is useful for is calibration. If your plan involves noticing a token, thinking about it, and buying, the shape above is the thing you are betting against. And if a scanner tells you a token looks clean, that is a statement about coordinated control and nothing else. It is not a floor under the price.