This page explains how a CryptoGrade letter grade is made. It is the reference document for our ratings, and it is meant to be specific enough to argue with.
If something here doesn't match what you see on a rating, that's a defect and we want to know.
The short version
Every asset we cover is scored on six weighted dimensions. That score then has to get past nine hard gates — pass/fail checks that sit outside the scoring. What comes out is a letter grade, published with a provenance record: which version of the rules produced it, which model ran it, and what the score was computed from.
Three ideas do most of the work here: weighted dimension, hard gate, and provenance record. Here they are in order.
1. The six weighted dimensions
A weighted dimension is one thing we measure, and a weight is how much it counts. A strong result on a heavily weighted dimension moves the grade more than a strong result on a light one.
These are ours, with the weight each one carries:
# | Dimension | Weight |
|---|---|---|
1 | Tokenomics & Value Accrual | 28% |
2 | On-chain Usage & Economics | 22% |
3 | Security, Decentralization & Durability | 18% |
4 | Ecosystem & Moat | 12% |
5 | Relative Valuation | 12% |
6 | Market Structure | 8% |
They add up to 100%. In plain terms, half the grade comes from two questions: does the token actually capture value, and is the network actually used.
Two things about these weights matter more than the numbers themselves.
They're published. Most rating providers name what they measure and keep the weighting proprietary. You can see ours, so you can disagree with ours. If you think usage should outweigh tokenomics, you now know precisely what you're disagreeing with.
They're fixed by rule version. The weights are not tuned per asset, and they are not adjusted after the fact to produce a preferred result. When they do change, the rule version changes with them, and that change is visible on every score produced afterwards.
2. The nine hard gates
A hard gate is not a score. It's a pass/fail check, evaluated separately, and tripping one forces the published grade to F no matter how well the weighted score came out.
Here's why we build it that way. Some problems shouldn't be averaged away. An asset can look good across every dimension we measure and still have one structural flaw that makes the average misleading. A weighted model on its own will quietly outvote that flaw — six decent scores drown out one bad one. A gate won't.
The nine:
Anonymous / unaccountable team — nobody identifiable stands behind the project.
Unaudited or previously-exploited contracts — the code has not been reviewed, or it has already been broken.
Unsustainable / Ponzi-like yield — returns that appear to be paid out of new deposits rather than out of anything the system earns.
Low float + high FDV + imminent unlocks — very little of the supply trades today, the fully diluted value sits far above the traded value, and a large tranche is about to be released.
Supply concentrated in insider wallets — a small number of related holders control enough to move or exit the market.
Thin or wash-traded liquidity — the volume on screen doesn't represent real depth you could transact against.
Unregistered-security / regulatory-kill profile — features that resemble an unregistered security, or a regulatory exposure severe enough to end the project. This is a risk screen, not a legal determination, and nothing here is legal advice.
Custody that can freeze, mint, or seize — someone retains the power to freeze balances, mint new supply, or take assets.
History of rug / abandonment / exploit — the project or its team has done this before.
When a gate trips, it forces the grade to F — regardless of how the weighted dimensions scored. The gate sits outside the weighted math entirely; nothing on the dimension side can offset it.
How to read a gate that passed
A passing gate means one thing precisely: no disqualifying condition was detected, under this rule version, on this basis. Read it that way — not as a clean bill of health. The basis label on the score, described below, is where you look to see what a given check rested on.
3. The provenance record
Every published score carries three fields.
Rule version. The exact version of the framework that produced the grade. Rules change; when they do, the version changes with them. This is what lets a grade be reconstructed later — you can tell whether it moved because the asset changed or because we changed.
Model identity. Which model performed the evaluation, recorded per score. Not "AI-powered" as a category. The identity itself, so you can see which grades were produced under which model.
Basis label. What the score was computed from. This is where input quality becomes visible: an input that was inferred, incomplete, or unavailable is labelled, rather than folded silently into an average.
This record is the part of the product we'd defend hardest. A grade without one is an opinion in a nice font. A grade with one is a claim you can check.
What the letters mean
Grade | Title | Score |
|---|---|---|
A | Prime | 80 – 100 |
B | Sound | 60 – 79 |
C | Speculative | 40 – 59 |
F | Failed | 0 – 39 |
Four bands, covering 0 to 100 with no gaps between them. There is no D or E — a score below 40 is an F.
A grade summarises the score and the gates. It is not a forecast, a signal, or a ranking of what will perform.
What we cover
We grade the top 200 assets, ranked by CryptoGrade's own score, highest to lowest — not by market cap. Anyone can see the top 10 without an account; the full 200 requires a free login.
What this framework does not do
Being clear about the boundary is part of the method.
It doesn't predict price. No dimension, gate, or grade encodes a view on what an asset will be worth. A grade assesses an asset against the rules on this page — nothing more.
It doesn't tell you what to do. CryptoGrade publishes ratings and research. It does not tell any reader to buy, sell, or hold anything, and a grade is not a recommendation or a signal.
It doesn't measure everything that matters. Six dimensions and nine gates are a deliberate simplification. Things that resist consistent measurement across 200 assets are out of scope, and their absence from the score is not evidence of their absence in the world.
It isn't right by construction. This is a set of rules applied by an automated process. Rules can be wrong, inputs can be stale, and outputs can be incomplete. That is what the provenance record is for.
When the rules change
Rule versions are how we change our mind in public. When a rule version changes and a grade moves because of it, the provenance record shows both the old and the new version — so a grade movement can be attributed to the framework or to the asset. Silent re-scoring would make every other claim on this page unverifiable, so we don't do it.