Assay

Does the coin do what it says?

Paste a Solana coin. Assay reads what it promises, tests each promise against the chain, its website and its code, and strikes a hallmark with the evidence for every mark.

3 free assays a day. Reading a report is always free, and a coin assayed in the last 12 hours opens its report at once.

Latest entry1 minutes ago

Jupiter $JUP

Governance token of Jupiter, a large live Solana DeFi platform; products verified, governance not shown

Read the report

Every coin Assay has tested, newest first.

Open the full ledger

Graded the way gold is.

Gold is rubbed on a black touchstone and its streak compared with needles of known purity. Assay does the same with a coin's claims. Each claim gets a mark, and the marks set the fineness: how many parts in a thousand of what it claims hold up.

The fineness is arithmetic on the marks, so the same marks always give the same grade. Claude marks the claims; nobody picks the number.

  • 99924 karatDoes what it says. 900 and up.
  • 75018 karatMostly does what it says. 750 and up.
  • 58514 karatSome of it works. 585 and up.
  • 3759 karatA little of it works. 375 and up.
  • GPGold platedMostly promises. 120 and up.
  • FeS₂PyriteThe claims don't hold up.
  • —UnmarkedClaims no utility. Nothing to test.

Verified

The evidence shows it working or true.

Counts in full.

Partly

Some of it is there; some of it isn't, or is smaller than claimed.

Counts 50%.

Unproven

Nothing Assay could read shows it, either way.

Counts 15%.

False

The evidence contradicts it.

Counts nothing.

Core claims weigh 3, features 2, details 1. If no core claim is verified, the fineness stops at 584. If a core claim is false, it stops at 374. A coin that claims no utility is left unmarked: there is nothing to test, and that is not a fail.

What happens in the minute after you paste.

  1. 1

    Collect

    Assay reads the coin's pump.fun listing and DexScreener pools, the mint and its biggest holders on Solana, where pump.fun sends its creator fees, its website and up to four of its pages, its GitHub, its X profile, every Solana address the site cites, and what CoinGecko and DefiLlama track for it.

  2. 2

    Read

    Claude Haiku 5.5 lists the utility the coin claims, quoting its own words and noting where each claim was made. Price talk and slogans aren't claims.

  3. 3

    Test

    Claude Opus 5.5 marks each claim against the evidence and names the facts every mark rests on. A page that only repeats a promise can't verify it.

  4. 4

    Strike

    The marks set the fineness. The report lists every claim, its mark, the evidence behind it, the risks Assay checks by rule, and what would raise the grade.

Holders assay all day.

Anyone gets 3 new assays a day. A wallet holding 100,000 $ASSAY or more gets 50. Connecting a wallet only shows Assay your address: no signature, no transaction.

$ASSAY hasn't launched. When it does, its mint goes on this page and in the whitepaper. Until then, any coin calling itself Assay's isn't ours.

Questions

Is a high grade a reason to buy?

No. Assay grades whether a coin's claims are true, not whether its price will rise. A 24-karat coin can still fall to zero, and an unmarked meme can still run.

Why is a meme coin “unmarked”?

Because it claims no utility, there's nothing to test. Unmarked means the coin is bought for what it is, not for what it does. Pyrite is for coins whose claims don't hold up.

Can the grade be wrong?

Yes. Claude can misread a page, and some things can't be checked from the outside. Every mark shows the evidence it rests on, so you can check it yourself. Anyone can ask for a fresh assay once a report is 12 hours old.

Can a project game it?

Writing more promises doesn't help: a claim is only verified by evidence other than its own words, like a program on chain, a working page or real code. Claims with nothing behind them pull the grade down.

What doesn't Assay check?

Price and charts, who the team are, X posts beyond the bio, Telegram, and pages that only draw themselves with JavaScript. It reads programs' on-chain facts, not their full code.