Methodology

What "the universe" means

The universe is the set of tokens that have passed the filters below at least once. Once a token enters, it never leaves — including when it dies. That is what makes the numbers meaningful.

Entry filters

Migrated to an AMM

The pair must trade on pumpswap, raydium, meteora or orca, against SOL, WSOL or USDC. Tokens still on a bonding curve do not qualify.

Market cap ≥ $500,000

Below this threshold there is more noise than signal.

Liquidity ≥ $50,000

As reported in USD by the exchange.

Age between 1 hour and 14 days

The first hour is skipped because the figures are meaningless then. After 14 days, tracking stops.

When a token counts as dead

Rugged

Liquidity fell below $5,000. Counts as a death.

Delisted

The exchange stopped returning the pair while its liquidity was still above the threshold. Counts as a death, but the cause is not known from the data.

Retired

The pair reached 14 days and observation ended. This is not counted as a death — the token is censored: it is removed from the denominator for every threshold beyond the point where it left observation. Counting it as a survivor instead would make long-horizon survival look artificially high.

The moment of death is not stored in the database — it is inferred from the pair's last minute-level snapshot, because snapshotting stops exactly then. The error is under one minute.

Why there is no survivorship bias

The classic mistake is to look at the tokens that are alive today and judge from them how likely a token is to survive. The dead ones have disappeared from that list, so the answer always comes out too optimistic.

Here the record is written at the moment of entry and never deleted. Every survival figure starts from all tokens that entered, not from the ones that made it. That is why every percentage on /stats is shown next to its denominator — so you can see how many observations it rests on.

The second requirement is subtler. A token that entered 3 hours ago says nothing about the 24-hour threshold: it has neither died nor survived that long. But if such tokens are simply dropped from the denominator, you get the opposite extreme — with a single dead token that lived 17 hours, the 24-hour threshold would have only that token left in the denominator and report 100% mortality.

So the estimate uses Kaplan-Meier: it is computed over intervals, and at each death the risk set contains only the tokens still under observation at that moment. Unfinished observations are censored rather than counted as survivors. The curve is defined only as far as the longest tracking period reaches — beyond that, /stats says "no data" instead of a number.

Data

Sources

DexScreener for discovery and minute-level snapshots, GeckoTerminal for minute candles. No API keys, no paid feeds.

Frequency

One snapshot per minute for every active pair; a search for new tokens every 5 minutes.

Current coverage

2026-08-15 20:55 UTC → 2026-08-16 22:59 UTC (26.1h), 7,064 snapshots, 15 tokens.

What this site does not do

It gives no trading signals and does not judge whether a token is good. It shows what happened to every token that passed the filter, and to how many of them.