Why do you score four criteria when other sites score ten?

Because we score what we can source. Six of the ten categories in a typical broker ranking — platform quality, mobile apps, research, education, customer service, execution — require using the product. We have not used these products. Publishing a number for them would present an impression as a measurement.

Can a broker pay to improve its score?

No, and the constraint is structural rather than a promise. The scoring engine imports the broker database and the weight profiles. It does not import the commercial register, which lives in a separate module. There is no code path from a commercial relationship to a score.

Why are sub-scores whole numbers?

Because the evidence is coarse. We are reading a published fee schedule and a regulator's entity record. That supports a judgment like 'this is a 7 and not an 8'; it does not support 8.3 versus 8.4. Whole-number sub-scores make the resolution of the underlying evidence visible rather than hiding it behind decimals.

What happens when a broker changes its fees?

The dataset carries one verification date per firm, shown on every row. We re-read the full dataset at least quarterly, and any individual firm within five business days of a credible report that its terms have changed. Corrections are logged under our corrections policy.

Methodology v1.1

How we score brokers

Everything on this page is a commitment you can hold us to. If a score on this site cannot be reproduced from the published sub-scores and the published weights, that is a bug, and our corrections policy covers it.

Dataset verified August 28, 202612 brokers scored1 withheld

The four rules

No unsourced figure is ever printed. Every number in our database is wrapped in a structure that requires a verification status and a source. A field we could not read is rendered as “not verified” — never blank, because a blank cell reads as zero or as “not offered”, and never estimated.
A score requires evidence a reader can check. The four scored criteria are built from documents anyone can pull up: published fee schedules and regulatory registers. Criteria that rest on hands-on assessment are not scored from marketing pages, and carry no weight until there is an assessment to publish.
Commercial relationships cannot reach a score. The scoring engine has no access to the commercial register. This is enforced by module boundaries, not by policy.
A ranking we cannot compute honestly is not published. Three rankings on this site are built, labelled as drafts, excluded from search and carry no table, because they depend on evidence we do not have.

Where we departed from the conventional weighting

The starting point for this methodology was a conventional nine-category framework: safety and regulation 25%, fees 20%, investment selection 15%, platform 10%, account types 10%, research and education 5%, customer service 5%, deposits and withdrawals 5%, transparency and execution 5%.

We could not source five of those nine. Platform, research, customer service and execution quality all require hands-on work with a funded account. Access and eligibility requires country-by-country onboarding terms. Rather than score them from marketing copy — which is what scoring them without testing amounts to — we set their weight to zero and redistributed it across the four pillars that rest on documents we actually read.

The trade is explicit: our rankings say less than a nine-category ranking does, and what they do say is checkable. We think that is the right trade for financial content, and it is the reason three rankings on this site remain drafts.

The four scored pillars

Safety and regulation

30%

Evidence: FINRA BrokerCheck entity record — legal name, CRD number, SEC file number and registration status — SIPC membership, and public enforcement actions from the regulator or SRO that brought them.

9–10: Decades of continuous registration under a single legal entity, member of the market's custody compensation scheme, and no headline retail enforcement matter identified in our sourced review.
7–8: Active registration and custody scheme membership, so custody protection matches the rest of the set, but a materially shorter registration record than the incumbents. Nothing here is a warning — it is less operating history to look at.
5–6: Active and covered, but carrying enforcement matters of real scale, or a recent entity migration a client would have to track.
0–4: Not a member of the market's custody scheme, or an inactive or lapsed registration. We would not publish a ranking place for a firm here without saying why on the page.

Fees and total cost

30%

Evidence: The firm's own published commission and fee schedule, read in full.

9–10: $0 on shares and ETFs with a genuinely cheap options schedule, no account service or inactivity fee, and little or nothing charged to move the account out.
8: $0 on shares and ETFs, but with one or two real costs the headline does not show: a mid-pack options fee, an exit charge, an OTC surcharge or a currency conversion markup.
6–7: A fee surface materially larger than the cheapest firms — an annual account fee unless waived, an expensive options schedule, transaction-fee funds, or a large exit charge.
0–5: Costs that would dominate a typical retail account's returns, or a schedule we could not read in full.

Investment selection

22%

Evidence: The asset classes named on the firm's own published pricing or product pages.

10: Every asset class we track, published on one schedule.
8–9: Equities, funds, bonds and options, plus either derivatives or genuine international market access. One or two classes missing.
6–7: A working range with one structural gap — typically no pooled funds, or no derivatives — that rules the firm out for a common strategy.
0–5: Equities and a little else. No pooled funds, no bonds and no derivatives: a narrow platform rather than a general one.

Account types and features

18%

Evidence: The account types named in the firm's own published account lineup.

9–10: Taxable, joint, every IRA variant, custodial and trust, plus at least one of a solo 401(k) or an HSA.
7–8: A full retail lineup including all IRA variants, custodial and trust. No HSA.
5–6: Taxable and IRAs in one person's name, but no joint, custodial or trust account — which settles it for anyone investing with a partner or for a child.
0–4: Individual taxable accounts only. No retirement wrapper at all.

The bands above are new in v1.1, and they are descriptive rather than aspirational: each was written against the scores already published, so nothing was re-scored to fit them. They exist because a rationale makes one score arguable while leaving the scale itself unfalsifiable — you could check that Fidelity charges $0.65 a contract without being able to check that a 9 for cost meant the same thing at Fidelity as at Webull. A build check fails if any published score falls outside its band.

Scores are not comparable across markets. The bands above are the United States rubric. Safety and selection ask the same question of any market and only change which register is read; cost and account types do not transfer at all. A market where trading shares is free and one where custody costs a percentage of the portfolio every year cannot share a cost scale, and an ISA is not a Roth IRA. Every market we publish carries its own bands, and a ranking table may only ever hold one market.

Weights shown are the default profile. Individual rankings re-weight the same four pillars — our low-cost ranking weights cost at 55%, our retirement ranking weights account types at 38% — and each ranking page publishes its own weights and the reasoning for them.

The five unscored pillars

These carry zero weight. Each entry states exactly what evidence would be required to start scoring it, so you can hold us to it and so we cannot quietly lower the bar later.

Access and eligibility0% — unscored: Would require: Country-by-country onboarding terms confirmed against each firm's own account-opening documents.
Platform and usability0% — unscored: Would require: Hands-on use of a funded account on desktop, web and mobile.
Research and education0% — unscored: Would require: Logged-in access to each firm's research suite and education library.
Customer service0% — unscored: Would require: Timed contact tests across phone, chat and email, repeated over several weeks.
Execution quality0% — unscored: Would require: Analysis of SEC Rule 605 and Rule 606 order-execution and routing disclosures across a full quarter.

How a score is calculated

Each pillar receives an integer from 0 to 10 with a written reason attached. The total is the weighted mean of those integers, expressed out of 10 and rounded to one decimal place:

total = Σ(sub-score × weight) ÷ Σ(weight), rounded to one decimal

A broker missing any weighted sub-score is not scored on a partial set — it returns no score at all and drops out of that ranking. Ties break on safety first, then on cost. Nothing commercial is ever a tiebreaker.

Alongside the number we show a band — Strong, Solid, Mixed or Limited — because the gap between 8.2 and 8.4 is not meaningful when the inputs are whole numbers, and a band stops readers over-reading a decimal.

Reproducing our scores yourself

Every ranking page has an Every sub-score and why section listing each pillar integer, its weight, and its contribution to the total. Take the integers, apply the published weights, and you will get the published total. If you do not, tell us — that is a reportable error under our corrections policy.

When we withhold

Two mechanisms keep unfinished work out of the search index. A withheld broker is a firm we confirmed on FINRA BrokerCheck but whose fee schedule we could not read from a primary source; it appears in no ranking, and the reason is published on its own page. A draft ranking is one that depends on an unscored pillar; the page is built and explains what is missing, but carries no table and is noindexed.

At this revision that is 1 withheld broker (SoFi Invest) and 3 draft rankings, against 12 scored brokers and 14 published rankings.

How often we re-verify

The full dataset is re-read end to end at least quarterly
Any individual firm is re-read within five business days of a credible report that its terms have changed
Entity records are re-checked against FINRA at every full re-verification
Every page shows the verification date for the data on it — we do not refresh a “last updated” date without re-reading the source

Questions about the methodology

Why do you score four criteria when other sites score ten?: Because we score what we can source. Six of the ten categories in a typical broker ranking — platform quality, mobile apps, research, education, customer service, execution — require using the product. We have not used these products. Publishing a number for them would present an impression as a measurement.
Can a broker pay to improve its score?: No, and the constraint is structural rather than a promise. The scoring engine imports the broker database and the weight profiles. It does not import the commercial register, which lives in a separate module. There is no code path from a commercial relationship to a score.
Why are sub-scores whole numbers?: Because the evidence is coarse. We are reading a published fee schedule and a regulator's entity record. That supports a judgment like 'this is a 7 and not an 8'; it does not support 8.3 versus 8.4. Whole-number sub-scores make the resolution of the underlying evidence visible rather than hiding it behind decimals.
What happens when a broker changes its fees?: The dataset carries one verification date per firm, shown on every row. We re-read the full dataset at least quarterly, and any individual firm within five business days of a credible report that its terms have changed. Corrections are logged under our corrections policy.