Conviction score
Every token we surface carries a 0–100 conviction score and a tier. The score answers one question and one question only: how much verifiable substance is there behind this token at the moment we looked?
It is not a price prediction. Nothing in it can be.
The tiers
Strong on several independent axes at once. Rare by construction.
Real signals with visible gaps. The bulk of what is worth reading.
Something is there and it is thin. Watch, do not chase.
Did not clear the floor. Scored, stored, never sent.
Tokens below 30 are scored and stored, never sent. You will not see them, but they exist in our data — which is how we can tell you what the tiers are worth rather than what we hope they are worth.
GOLD is rare on purpose and it is meant to stay rare. It requires strength on several independent axes at once, and most real tokens are strong on one or two. If GOLD started arriving daily, the honest response would be to check what broke — not to celebrate.
The six channels
The score is the sum of six independent checks. Each has a ceiling, and the ceilings are not equal — they encode how much we think each channel is actually worth knowing.
Is anyone actually building anything?
A reachable GitHub, whether that repo predates the token, whether it is written in a language a real project would use, plus a whitepaper, docs and an audit.
Vaporware is cheap to fake in a bio and expensive to fake in a commit history.
Can you get your money back out?
The measured round-trip cost of a simulated buy→sell, the tax curve shape, liquidity burned or locked, wash-trade result, and source verification.
This is the only channel where a bad enough answer also triggers a penalty against the total.
Is the attention real?
Accounts above 10K and above 50K followers talking about it, how many distinct authors, and an organic-versus-coordinated read.
Paid-signal bots are the cheapest fake in crypto, so reach is counted by tier and coordination is counted against.
Is there a coherent reason this exists right now?
Narrative strength, originality, and whether it attaches to a real event, a culture or nothing at all.
A token with no story is not necessarily bad. A token whose story is copied from last week's winner usually is.
Who made this, and what happened last time?
Wallet age, funding source, prior launches that ran, and prior launches that rugged.
A prior rug is the single strongest negative in the channel — it subtracts, it does not merely fail to add.
How few hands hold it?
The share of supply in the top ten non-contract wallets, scaled continuously rather than banded.
Small ceiling on purpose. A one-hour-old token is always concentrated; punishing that hard would punish being early.
Then the penalties, which subtract from the total
- -30The sell simulation classified the curve as a honeypot, or the pool hook belongs to a known honeypot family.
- -25A measured round-trip cost of 50% or more. Sellable in principle, ruinous in practice.
- -20Confirmed wash trading — the seller does not receive the proceeds of their own sell.
- -15A conditional tax pattern: the contract charges differently depending on who is calling.
Penalties are capped at -50 points combined. A token can therefore score well on five channels and still land in SKIP — which is the intended behaviour, not an edge case.
Why six, and why these six
A single-number score is trivially gameable. Buy followers, wash some volume, and a tool that reads one dimension calls you legitimate.
Six independent dimensions force a token to be real on several fronts at the same time. A beautiful narrative with no repository gets caught by Codebase. Enormous "hype" from three hundred accounts with four hundred followers each gets caught by Social. A clean, verified contract deployed by a wallet with two prior rugs gets caught by Deployer.
The ceilings encode a judgement you can argue with:
- Codebase and Security carry 25 each — half the total between them. Both are expensive to fake and cheap for us to verify.
- Holders carries 5. Concentration is real information, but a token an hour old is always concentrated. A big holder ceiling would mostly punish being early, which is the opposite of what we want.
- Deployer carries 10, and it is the only channel where a single fact can take away more than it gives. A prior rug subtracts.
What a score cannot survive
Four findings do not merely fail to add points — they subtract from the total. They are the difference between "we found little evidence" and "we found evidence of harm".
This is why a token can score respectably on five channels and still land in SKIP. That is the intended behaviour: a honeypot with a real GitHub is still a honeypot.
Reading a score honestly
A high score is a statement about evidence, not about price. GOLD means the token cleared a demanding bar of due diligence at the moment we measured. Tokens that clear it still go to zero, regularly. Anyone telling you otherwise is selling something.
A low score is often a statement about our coverage, not the token. Absence of evidence is not evidence of absence: a project whose GitHub we could not reach scores the same on Codebase as one that has no GitHub. Where we could not measure something, the card says so.
The score is a snapshot. It is computed from what was verifiable at alert time. A contract that can be upgraded (see the safety stack) puts a shelf life on every number in it.
➡️ Next: Trade labels — the momentum axis, which answers a completely different question · What we publish and what we don't
Anything on this page unclear, or wrong?
Tell us →