How we rate things

What makes a rating real

Three things separate a rating from a listing farm, and all three are testable by you, the reader, on any page we publish.

• Where the numbers come from. Not the provider’s own website — our own measurement, with the date and time we took it. Anything we take from a provider’s price list is labelled as published, not measured. We never blend the two silently.
• How the score is built. Criterion → weight → score → the reason for that score. Never “we think”.
• What is missing. Every entry has a “where we docked points” section and a “who this is not for” section. Without those two, a card is an advertisement.

The scale

Every criterion is scored 0–10. The total is the sum of score × weight, and the weights inside a category always add up to 100. One rating uses letter grades instead, because what it assesses is a public record rather than a number — it has its own section below.

The seven universal criteria

These recur in nearly every category; only the weight changes. That is what makes ratings comparable across categories instead of being seven unrelated opinions.

Evidence levels — and what we refuse to fake

This is the part most rating sites leave out, because it is the part that costs something. We split evidence into three levels by what it takes to obtain.

What this means in practice — the rule that governs everything else. If we cannot verify a criterion, we delete it. We do not carry it as an unscored placeholder, we do not let it quietly dilute the weights, and we absolutely do not fill it with a plausible guess. Every criterion you see published is either measured by us or sourced to a primary document, which means a score always rests on 100% of the stated methodology — never on a fraction of it with an asterisk.

The clearest example: our exchange ranking has no support criterion. Judging support honestly means opening real tickets as a paying customer and timing the replies, on a weekday and at a weekend. We have not done that. So rather than score it from impressions, or leave a 5% hole in the weights, we removed it entirely. When someone does the work, it comes back — and the changelog will say so on the day it does.

We will not manufacture a TXID, a support response time, or a field test we did not run. A fabricated measurement is worse than a missing one, because it is more convincing.

Where that rule bites, in what you are reading today

It is easy to state a principle like that and quietly not apply it. So here is exactly where it cost us something in this release:

• We have not bought the hardware wallets. That is Level C. So the hardware-wallet rating scores the chip and its certificate number, the licence and the reproducibility of the firmware builds, the audits, and whether the seed can be restored into a rival’s device — all of which are documented and checkable. It does not claim we plugged the devices in. Where a criterion would have required our hands on the buttons, it is marked as read from documentation, not as observed.
• The exchanger rating measures rates, not delivery times. A quote can be requested for free without executing anything, so we did that, for every service, in the same window. Transfer speed and support response cannot be had without sending real money, so those criteria were deleted rather than guessed — not marked “n/a”, deleted, with the remaining weights summing to 100 on their own.
• A take rate is not a fee schedule. Where we divide thirty days of fees by thirty days of volume — for decentralised exchanges and for trading terminals — the result is what users paid on average across the pools and routes they actually used. It is the most honest number available from public data, and it is still an average: your individual trade can sit either side of it, and we say so on every page that carries one.
• The crypto-card rating scores what public terms reveal, not a field test. The true conversion spread is only visible after you load a card and spend from it — money we do not get back — so we do not score that. We rank on the reward you actually keep, on token risk priced against real prices, on who issues the card and whether an American can get it: the crypto-card rating.

Hard disqualifiers

Fail any of these and the service is excluded from the ranking, regardless of its scores, its fees or how much traffic it would bring us.

• User funds never returned after an incident.
• A pattern (not a single complaint) of obstructed withdrawals.
• A custodial service with neither proof-of-reserves nor a licence.
• Volume falsification, per independent metrics.
• An anonymous team running a custodial model.
• Operating without a required licence in the country a given cut is written for.

We publish the exclusions on the page itself — who was considered, who was cut, and why. It is usually the most-cited part of a rating, because it is the part nobody else is free to write.

Letter grades: DAO treasury & liquidity managers

Firms that manage a DAO’s treasury or a token’s liquidity mostly publish nothing we could measure: no fee we can test, no book we can walk. What can be established is what the public record shows. So this rating grades three blocks from A to F instead of scoring criteria from 0 to 10, and a grade is only ever as good as the sources behind it.

Who is included: organisations that manage other parties’ assets or liquidity under an agreement with a DAO or a token issuer — treasury diversification, protocol-owned liquidity, market making, managing AMM positions. Who is not: custodians, multisig tooling and exchanges.

• B and D are given when the condition for the grade above is partly met.
• For sorting, A = 4, B = 3, C = 2, D = 1, F = 0. The overall grade is the average of the three blocks, rounded to the nearest letter, halves up. Only the letter is shown.
• Two firms with the same average are ordered by their mandate grade, then alphabetically.
• Under every firm is a one-sentence verdict, written from the facts on its card and signed by a member of the desk.

Every fact behind a grade carries its source and the date we read it. A fact we could not source is shown as “not published” and graded as missing; anything better than an F has to be backed by at least one source. A firm we could not verify is not published at all. The arithmetic — letters to points, the average, the rounding and the tie-break — is code with a test, not a judgement made per firm.

No affiliate links, and why that is the whole point

This section is not monetised. No affiliate links, no paid placement, no sponsored positions. That is not modesty — it is the only real competitive advantage available, because the rest of the crypto-rating market is monetised through referral commissions and is therefore distorted in the same predictable direction.

Not being paid is what makes these sentences possible:

• “You do not need this category of product at all.”
• “Take the free open-source option; the paid one is worse.”
• “There will be no ranking here — the entire category is toxic, and here is why.”
• “First place goes to the service with no referral programme.”

A site paid per signup cannot write any of those, even if it wants to. See how this is funded for the full policy.

Dates, not vibes

Every number carries the moment it was captured, not a vague “updated recently” at the top of the page. An order-book spread measured last month tells you almost nothing about today, and pretending otherwise is how rating pages quietly become fiction.