How we score

Brokers

• Regulation & licensing

  27%

  The best licence a broker holds, plus a bounded bonus for holding several serious ones — counting only the companies that would open a retail account. A group licence held by an arm that deals with other firms is a fact about the group, not a protection for you, and it scores nothing here. Tier A regulators run a statutory compensation scheme and a public register; tier C is registration only.

• Regulatory & legal record

  10%

  What regulators, prosecutors and courts have actually done about the company, on the record and with the document behind each one. A prosecution costs most, then a restriction, a fine, a public warning, and last a private claim — anyone may sue anyone. An allegation costs four fifths of what the same thing costs once decided, because a regulator bringing a case is information and pretending otherwise would let a firm under active prosecution score as though nothing were happening. Matters fade after three years and stop counting after ten. Crucially, a broker nobody here has searched is excluded rather than given ten: no data is not a clean record.

• Published trading cost

  18%

  Published EUR/USD spread and round-turn commission reduced to a single figure. A $7 commission per standard lot is worth about 0.7 pips, so the two are directly comparable.

• Payments & withdrawals

  18%

  Breadth of funding methods, the broker’s own stated withdrawal processing time, and how much it takes to open an account.

• Platforms & execution

  13%

  How many platforms are offered, the execution model, and whether copy trading is built in.

• Verified reviews

  9%

  Mean of verified reviews. Counted only once a broker has at least five — below that it is excluded rather than guessed at.

• Corporate transparency

  5%

  Four disclosures: entity mapping, audited accounts, segregated client funds, public ownership.

Prop firms

• Rule fairness

  27%

  How drawdown is measured carries more than a third of this component on its own. Static drawdown is fixed against your starting balance; trailing drawdown follows equity upward, so an unrealised spike permanently raises the floor. The rest is headroom, profit target, deadline, and whether a consistency rule, news ban or weekend ban applies.

• Payout terms

  22%

  Profit split, how often a payout can be requested, and how long after funding the first one becomes available.

• Challenge cost

  18%

  Challenge fee normalised to a $100k account. Firms price many account sizes, so a headline fee compares nothing.

• Platforms & markets

  13%

  How many platforms are offered and how many markets can be traded.

• Transparency

  10%

  Three disclosures: a published rule-change history, the legal entity behind the firm, and the broker executing the trades.

• Evidence behind this record

  10%

  How much of this record anybody has been able to confirm, which is a statement about us rather than about the firm. Four points for a firm somebody has researched and written up at all; three more if the firm’s own pages answered our requests, so its rules were read where they are published rather than where a review site repeated them; three more if an independent document — a company register or a statutory filing — names the companies behind it. Excluded, not zeroed, where nobody has done the research. A firm that blocks automated readers is not being dishonest, and this does not say it is; it says how far a reader should trust the rest of the row.

Exchanges

• Solvency evidence

  25%

  What evidence exists that customer funds are there. A self-published proof of reserves is real but the weakest kind — unaudited, chosen by the exchange, silent on liabilities. An audit by a named third party, or the continuous disclosure a public listing forces, scores higher.

• Security record

  20%

  Years since the last customer-funds breach, softened by whether users were made whole. An exchange that was hacked and covered every loss is not in the same category as one that was not.

• Regulatory and legal record

  15%

  The same register the brokers are scored against, read for exchanges. It carries more weight here — 0.15 against 0.10 — because three of the eight ranked have pleaded guilty to a US federal offence or paid a penalty in the billions, and a model that ranked them on fee tiers while saying nothing about that would be answering a question nobody asked. A case brought and then dismissed costs nothing at all: the SEC sued Coinbase and Kraken and dropped both with prejudice and no penalty, and a register that recorded the accusation and not the ending would be a rumour column with dates. An exchange nobody has searched is excluded rather than given ten.

• Trading fees

  14%

  The taker fee actually charged at the lowest tier.

• Liquidity

  10%

  Reported spot volume, on a logarithmic scale and used only as a band. Volume is self-reported and has been inflated industry-wide for years, which is why it carries the least weight here.

• Transparency

  8%

  Three disclosures: a public fee schedule, the legal entity, and incident reports.

• Evidence behind this record

  8%

  How much of this record anybody has been able to confirm, on the same rule the prop firms use: four points for a record somebody has researched and written up, three more if the company’s own pages answered our requests, three more if an outside document — a regulator’s notice, a court filing, a company register — names it. Excluded rather than zeroed where nobody has done the research. It carries less weight here than on the prop side because these companies are unusually well documented by other people: a guilty plea is a stronger record than anything a firm publishes about itself.

Memecoins

• Deployer control

  45%

  What the deployer can still do after launch: print supply, freeze balances, rewrite balances outright, change metadata, or tax transfers. Two findings cap the score no matter what else is true -- a blocked sell path and a mutable balance authority. A name carrying invisible bidi characters caps it too, because a name crafted to render as something it is not is a deliberate act.

• Exit liquidity

  25%

  Money actually sitting in the pool, on a logarithmic scale. It is the practical measure of whether you can exit at size.

• Transfer tax

  15%

  Buy and sell tax on EVM chains; on Solana, whether a transfer fee or transfer hook is attached.

• Trading activity

  15%

  Whether trading is genuinely two-sided. A book that is all buys and no sells has not been tested for exit.

Reviews are the most attacked surface a ranking site has: the company wants good ones, its competitors want bad ones, and from the server’s side both look exactly like a real customer. So publishing and counting are separate acts. A review is live the moment it is written, marked unverified, and reaches a score only once an editor has checked the evidence behind it — which means buying a hundred five-star reviews buys a hundred unverified paragraphs and moves nothing.

Below 5 checked reviews the component is excluded rather than scored low, like every other component here. Both averages are shown — the checked one the score uses, and the one covering everything anyone wrote — because showing only the first hides what people are saying and showing only the second hands the score to whoever writes most.

Reviews move the broker ranking and no other. The prop firm and exchange models were published without a reviews component, and adding one means changing weights that are already public — a decision to make openly, not a side effect of shipping a feature. Until then their reviews are read and not counted, and every one of those pages says so.

The broker status signal on each page is unverified by design — that is what makes it fast enough to be useful during an outage. It is counted by distinct reporters over a rolling window, with the count and the threshold published so you can weigh it yourself.

It never enters a score. Scores move only on facts a person checked against a primary source and recorded with that source. Mixing an unverified crowd signal into a number we ask readers to trust would quietly destroy the distinction the rest of this page is built on.

Every upstream is free and unauthenticated: CoinGecko for coin prices, GeckoTerminal for new pools, GoPlus for contract audits. Nothing here depends on a paid plan, which is a deliberate constraint — it keeps the cost of running this site near zero and means no ranking can quietly become a function of what we could afford to license.

Free tiers go quiet. When one does, the affected page says so plainly rather than serving figures we could not confirm, and every live page carries the timestamp of the fetch behind it.

Every regulator publishes a register precisely so anyone can check a firm before dealing with it. Not every one of them can be read by a machine in a datacentre: some need a registered key, some render their search in the browser, some refuse our address outright. 12 of the 40 licences on this site are compared against the issuing regulator’s own register once a day. The rest are not, and every page that carries one says so rather than letting it pass as checked.

• CySEC · 8 licencesCyprus Securities and Exchange Commission — read once a daychecked
• ASIC · 4 licencesAustralian Securities & Investments Commission — read once a daychecked
• FCA · 5 licencesAPI requires a registered key; the public search page renders client-sidenot machine-checked
• FSC-MU · 4 licencesrefuses requests from datacentre addresses (HTTP 403)not machine-checked
• FSCA · 3 licencesno reader written for this register yetnot machine-checked
• FSA-SC · 3 licencesno reader written for this register yetnot machine-checked
• FSA-SVG · 3 licencesno reader written for this register yetnot machine-checked
• BaFin · 2 licencesno reader written for this register yetnot machine-checked
• DFSA · 2 licencesno reader written for this register yetnot machine-checked
• SCB-BS · 2 licencesno reader written for this register yetnot machine-checked
• IFSC-BZ · 2 licencesno reader written for this register yetnot machine-checked
• CMA-KE · 1 licenceno reader written for this register yetnot machine-checked
• MISA · 1 licenceno reader written for this register yetnot machine-checked

Dates come from the institution that sets them — the BLS release schedule, the Fed’s FOMC calendar, the ECB’s Governing Council calendar. Nothing is copied from another calendar site and nothing is inferred from a pattern. A release time is shown only where the source states one: the BLS publishes exact times and the zone they are in, the Fed and the ECB publish dates without them, and a guessed time on a rate decision would be the most quietly harmful number on the site.

High impact

employment situation · consumer price index · rate decision

Medium impact

producer price index · job openings and labor turnover · employment cost index · import and export price indexes · productivity and costs · real earnings

Low impact

everything else — regional, sectoral or backward-looking

The institutions do not rank their own releases, so that grading is ours. It is a list of names rather than a model, deliberately: a list can be argued with, and if you think a release is in the wrong band you can see exactly what to argue about.

Publishing live spreads honestly means running a terminal on every broker, around the clock, and being accountable when it silently stops. We do not run that, so we do not claim it. Instead each broker’s published cost is shown with the date a human last checked it against the broker’s own pages, and anything unverified is labelled as such on the page. A figure you can trace beats a figure that merely looks live.

A broker with no verified reviews yet is not the same thing as a broker with terrible reviews. When a component has no data, it is dropped and the remaining weights are renormalised so they still sum to one. Every broker page lists which components were excluded, so a score is always readable against what actually went into it.

1. 17 September 2026

   Prop firms: added the evidence behind a record at 0.10, taken from rules, payout and cost.

   Reading all eight prop records back against their sources produced a ranking we could not defend. Seven had a figure wrong and the corrections went one way — four firms had published a profit split that was the ceiling of a range rather than what a newly funded trader is paid. But four firms answer our requests with 403 or 429, so their figures could not be checked at all, and they finished at the top: the firm nobody could verify ranked first. A model that lets "we could not look" beat "we looked and it was worse than we thought" is measuring the wrong thing. The confidence behind a record is now a component with a published weight instead of a caveat nobody reads. Rule fairness went from 0.30 to 0.27, payout terms from 0.25 to 0.22, challenge cost from 0.20 to 0.18.

2. 16 September 2026

   Added the regulatory and legal record at 0.10, taken from regulation, platform and reviews.

   The model read who supervises a broker and never what any of them had caught it doing. A regulator suspending the voting rights of a broker’s controlling owner, and a financial crime agency attaching his assets and prosecuting the company, moved this ranking by nothing at all — that broker sat fifth of ten on spread and withdrawal speed while both were true. Regulation went from 0.30 to 0.27, platform from 0.15 to 0.13, reviews from 0.10 to 0.09.

3. 16 September 2026

   Regulation counts only the companies that would open a retail account.

   A broker was leading with an FCA licence held by a company whose own filed accounts describe a B2B business. A licence nobody reading this site could be a client of is a fact about the group, not a protection for the reader, and it now scores nothing.